Partially Hyperbolic Dynamics in Dimension Three PDF

c Cambridge University Press, 2017
Ergod. Th. & Dynam. Sys. (2018), 38, 2801–2837

doi:10.1017/etds.2016.142
SURVEY
Partially hyperbolic dynamics in dimension
three
PABLO D. CARRASCO†, FEDERICO RODRIGUEZ-HERTZ‡,
JANA RODRIGUEZ-HERTZ§ and RAÚL URES§
† ICMC-USP, Avenida Trabalhador São-carlense 400, São Carlos, SP 13566-590, Brazil
(e-mail: pdcarrasco@gmail.com)
‡ PSU Mathematics Department, University Park, State College, PA 16802, USA
(e-mail: hertz@math.psu.edu)
§ IMERL-FING, Julio Herrera y Reissig 565, Montevideo 11300, Uruguay
(e-mail: jana@fing.edu.uy, ures@fing.edu.uy)
(Received 5 December 2016 and accepted in revised form 5 December 2016)
Abstract. Partial hyperbolicity appeared in the 1960s as a natural generalization of

hyperbolicity. In the last 20 years, there has been great activity in this area. Here we survey
the state of the art in some related topics, focusing especially on partial hyperbolicity in
dimension three. The reason for this is not only that it is the smallest dimension in which
non-degenerate partial hyperbolicity can occur, but also that the topology of 3-manifolds
influences the dynamics in revealing ways.
1. Introduction
Partial hyperbolicity was introduced in the late 1960s as a generalization of the classical
notion of hyperbolicity. In hyperbolic systems, the tangent bundle splits into two directions
that are invariant under the derivative: one, the stable direction, is contracted, and the
other, the unstable direction, is expanded. More precisely, a diffeomorphism of a compact
manifold f : M → M is hyperbolic if the tangent bundle splits as T M = E s ⊕ E u , where
D f (x)E xs = E sf (x) and D f (x)E xu = E uf (x) and, for each pair of unit vectors v s ∈ E xs and
v u ∈ E xu ,
kD f (x)v s k < 1 < kD f (x)v u k.
The simplest examples of this behavior are the hyperbolic (also known as Anosov)
automorphisms on tori.
Partially hyperbolic diffeomorphisms, in turn, allow one extra, center direction, which
is neither as expanded as the unstable one nor as contracted as the stable one. Again the
2802 P. D. Carrasco et al
simplest examples are certain automorphisms of the torus, for instance

 
210
A = 1 1 0 .
001
This matrix has three eigenvalues: λ > 1, λ−1 and 1. Thus, the associated automorphism
of T3 has three invariant bundles parallel to its three eigendirections. Two of these bundles,
the ones associated to λ and λ−1 , have a hyperbolic behavior and the one associated to 1
corresponds to the center direction.
Another classical (and more interesting) example in dimension three comes from the
action of the diagonal subgroup on quotients of G = PSL(2, R). Consider a left invariant
Riemannian metric on G.
1/2 0

Statement. The right multiplication by d1 = e induces a partially hyperbolic
0 e−1/2
diffeomorphism of G, ψ := Rd1 : G → G, ψ(g) := Rd1 (g) = gd1 .
Let t/2
1t 10 e 0
−
ut = , ut =+
, dt =
01 t1 0 e−t/2
be, respectively, the stable horocycle, the unstable horocycle and the diagonal 1-parameter
groups. Consider the foliations generated by these 1-parameter groups
F s (g) = u −t g; t ∈ R , F u (g) = u +t g; t ∈ R , F c (g) = {dt g; t ∈ R} .

ψ, respectively, contracts, expands and is an isometry on the leaves of these foliations.

Notice that
s dt = dt u e±t s dt .
±
u± (1.1)
This equation shows that ψ intertwines leaves of the foliation F s and, similarly, with F u .
Moreover, ψ preserves the leaves of F c . We also get from equation (1.1) that
ψ(gu ±
s ) = gu s d1 = gd1 u e±1 s = ψ(g)u e±1 s .
± ± ±
Remember that we choose a left invariant metric on G, so, using its distance function, we
obtain that
s ), ψ(g)) = d(gd1 u e±1 s , gd1 ) = d(u e±1 s , e) = d(gu e±1 s , g)

± ± ±
d(ψ(gu ± (1.2)
and
d(ψ(gdt ), ψ(g)) = d(gdt d1 , gd1 ) = d(gd1 dt , gd1 ) = d(dt , e) = d(gdt , g). (1.3)
From equation (1.2), we obtain that the leaves of Fs
are exponentially contracted in the
future and the leaves of F are exponentially contracted in the past by ψ. Equation (1.3)
u
shows that ψ acts isometrically on the leaves of F c . Define E gσ = Tg F σ , σ = s, u, c and

observe that we obtain a splitting
E gs ⊕ E gu ⊕ E c = Tg G,
where the first direction is exponentially contracted, the second is exponentially expanded
and the third one is isometric.
Partially hyperbolic dynamics in dimension three 2803
Let us give a different description of the invariant bundles. Let us describe first what
would be the partially hyperbolic splitting for the tangent space to the identity element
e = 10 01 . The tangent space to the identity element Te G is naturally identified with the
Lie algebra of G, g = sl2 . On g, we have three distinguished elements
1
01 + 00 0
U =−
, U , D= 2 .
00 10 0 − 12
Observe that
[D, U + ] = U + , [D, U − ] = −U − , [U + , U − ] = 2D. (1.4)
Let E eu = span{U + }, E es = span{U − } and E ec = span{D}. Clearly, U + , U − , D are

linearly independent and hence
E es ⊕ E eu ⊕ E ec = g = Te G.
Given g ∈ G, let us define Rg : G → G by Rg (h) = hg. Let De Rg : Te G → Tg G be the

derivative of Rg over the identity and define E gσ = De Rg (E eσ ), σ = s, c, u. Let us show
that these bundles are, accordingly, the unstable, stable and center bundle. To this end, we
shall use L g : G → G, L g (h) = gh, the left translation. Since the Riemannian metric we
choose is a left invariant metric, we get that L g is an isometry for every g. Observe that
(L g )−1 = L g−1 .
Both ψ and the right group actions remain well defined if we take a quotient of
PSL(2, R) under the left action of a lattice 0 and, since the Riemannian metric we
choose is left invariant, we obtain a Riemannian metric on the quotient 0\PSL(2, R).
The contraction and expansion properties are hence preserved on this quotient also. Recall
that the 3-manifold 0\PSL(2, R) can be naturally identified with the unit tangent bundle
of a closed surface (compact in the case when 0 is co-compact) and ψ with the time-one
map of the geodesic flow of a metric of constant negative curvature on this surface (again
the extra direction is given by the direction of the diagonal flow); see [KH95].
Precisely, we say that a diffeomorphism f of a closed manifold M is partially

hyperbolic if the tangent bundle splits as T M = E s ⊕ E c ⊕ E u , where D f (x)E xu = E uf (x) ,
D f (x)E xc = E cf (x) and D f (x)E xs = E sf (x) and if, for each unit vector v s ∈ E xs , v c ∈ E xc and
v u ∈ E xu ,
kD f (x)v s k < 1 < kD f (x)v u k and (1.5)

kD f (x)v k < kD f (x)v k < kD f (x)v k.
s c u
(1.6)
The set of partially hyperbolic diffeomorphisms is C 1 -open in Diff1 (M) (see, for
instance, Theorem 2.15 in [HPS77]. In other words, a C 1 -perturbation of a partially
hyperbolic diffeomorphism is partially hyperbolic.
1.1. More examples. The examples given in the Introduction fall into more general
classes.
1.1.1. Time-one maps of Anosov flows. Consider an Anosov flow in a 3-manifold φt :

M → M such that the tangent bundle of M splits into three sub-bundles invariant under
Dφt : T M = E s ⊕ X ⊕ E u , where X is the direction tangential to the flow, and such that,
for each unit vector v s ∈ E s and v u ∈ E u ,
kDφt v s k < 1 < kDφt v u k.
Then the time-one map of the flow φt is a partially hyperbolic diffeomorphism (exercise).
Examples of this kind of partially hyperbolic diffeomorphism are the ones induced by
the diagonal action on PSL(2, R) mentioned above.
As a different type of example, consider the suspension of an Anosov map A : T2 → T2
by the constant roof function one. Let A be a hyperbolic automorphism on T2 and consider
in T2 × R the equivalence relation (x, t + 1) ∼ (Ax, t). Then M = T2 × R|∼ is a smooth
manifold and f ([x, t]) = [x, t + 1] is a partially hyperbolic diffeomorphism.
We remark that these examples are truly different: for the second one the distribution
E u ⊕ E s is integrable whereas for the first one it is not.
Note that, in both examples, the corresponding Anosov flow is transitive, but there also
exist non-transitive Anosov flows [FW80a]. There is, in fact, a huge zoo of Anosov flows
(see, for instance, [Bar98, BL94, Fen94, Fri83, HT80]); so classification of time-one
maps of these flows is naturally a difficult task.
1.1.2. Skew products. Another example of a partially hyperbolic diffeomorphism is

a certain kind of skew product that is a circle extension over the 2-torus of the form
f (x, θ ) = (Bx, h(x, θ )) where B is a hyperbolic automorphism of the 2-torus and h(x, .)
are circle rotations. The resulting ambient manifold is a 3-nilmanifold. In the case when
the product is direct, the ambient manifold is the 3-torus. To this class belongs A, the
3-toral automorphism defined at the beginning of this introduction.
1.1.3. DA-diffeomorphisms. A DA (‘Derived from Anosov’) partially hyperbolic

diffeomorphism is one that is isotopic to an Anosov one. By a result of Franks,
DA-diffeomorphisms are semi-conjugate to Anosov diffeomorphisms. The prototypical
example in this class is Mañé’s [Mañ78]. It is obtained by taking a linear Anosov map in
T3 with eigenvalues λss < 1 < λu < λuu and making a bifurcation of the origin into three
points along the weak unstable direction (see Figure 1). The resulting map is a dynamically
coherent (robustly) transitive partially hyperbolic diffeomorphism.
2. Some open problems

Partial hyperbolicity appeared as a natural generalization of hyperbolicity. One main
motivation for studying partially hyperbolic systems was ergodicity. A system f : M →
M preserving a measure m is ergodic when, on average, any two events tend to be
independent: that is, for any two measurable sets A and B,
N −1
1 X
lim m( f n (A) ∩ B) = m(A)m(B).
n→∞ N
n=0
Ergodicity is therefore an interesting property from the physical point of view. Since
conservative hyperbolic systems are ergodic [AS67], it was asked whether adding some
F IGURE 1. Mañé’s DA partially hyperbolic diffeomorphism.
hyperbolicity to a system and then perturbing would generate ergodicity in the whole
system. Concretely, Pugh and Shub [PS97] asked whether, for any system, taking the
product of it by a sufficiently strong hyperbolic system (and hence obtaining a partially
hyperbolic system) and then making a small perturbation would yield ergodicity in a robust
way. They went further to state Conjecture 2.2 below. The area of partial hyperbolicity
has become very active since then, and other aspects have attracted interest as well. These
are also stated below.
2.1. Ergodicity. As stated above, one problem in partially hyperbolic dynamics

is studying the ergodicity of conservative systems; to be more precise, of C r -
diffeomorphisms preserving a smooth volume. We shall denote conservative systems by
Diffrm (M), where m denotes the probability measure arising from this smooth volume.
An equivalent definition of an ergodic system f is that any measurable set A satisfying
f (A) = A must also satisfy either m(A) = 1 or m(A) = 0. In other words, an ergodic
system is one not admitting an invariant measurable set with intermediate measure.
An example of a non-ergodic partially hyperbolic diffeomorphism is the toral
automorphism A defined at the beginning of the introduction. This automorphism can
be seen in the following way: A = B × id, where B is the automorphism on the 2-torus
generated by the matrix 21 11 and id is the identity on the circle. Indeed, any interval of the
circle times the 2-torus is an invariant set with intermediate measure.
One can easily perturb this system to obtain an ergodic partially hyperbolic one.
For example, g = B×(irrational rotation). However, this new system g can be easily
perturbed to obtain, again, non-ergodic diffeomorphisms. It is, in fact, approximated by
diffeomorphisms g 0 = B×(rational rotation), which are non-ergodic.
Pugh and Shub were the first to conjecture that ergodicity is, in fact, very abundant in
the partially hyperbolic world; a conjecture that remains open today (Conjecture 2.2). As
a matter of fact, this conjecture was made public for the first time in Montevideo, in 1995
[PS96]. We thank Burns for recalling this fact, and Mike Shub for confirming it. Pugh
and Shub considered not only ergodicity, but a stronger concept, as given by the following
definition.
Definition 2.1. A conservative C 2 diffeomorphism f : M → M is stably ergodic (in
Diff1m (M)) if there exists U ⊂ Diff1m (M), a neighborhood of f , such that every g ∈ U
of class C 2 is ergodic.
Until 1994, the only known examples of stably ergodic diffeomorphisms were Anosov
diffeomorphisms, that is, hyperbolic ones. In 1994, Grayson, Pugh and Shub found the first
non-hyperbolic examples. A year later, Pugh and Shub made the following conjecture.
C ONJECTURE 2.2. (Pugh and Shub (1995) [PS96]) Stable ergodicity is C r -dense among
volume-preserving partially hyperbolic diffeomorphisms, for all r > 1.
Pugh and Shub suggested a program in order to prove their conjecture. It involves
accessibility, as defined below.
Definition 2.3. Two points x and y are in the same accessibility class if there is a path that
is piecewise tangential to E s or E u joining them. The partially hyperbolic diffeomorphism
f has the accessibility property if there is only one accessibility class. The diffeomorphism
has the essential accessibility property if any measurable set that is a union of accessibility
classes has either full or null measure.
In the picture below, the points x, y and z are in the same accessibility class.
Exercise 2.4. Prove that the example of the action of the diagonal subgroup defined on
page 2802 has the accessibility property.
Obviously, accessibility implies essential accessibility, but the converse is not true.
Exercise 2.5. Show that the automorphism on T3 defined by

 
0 0 1
A =  0 1 −1 
−1 −1 5
has the essential accessibility property, but does not have the accessibility property.
C ONJECTURE 2.6. (Pugh and Shub) If f is a C r conservative diffeomorphism, with r > 1,
(essential) accessibility implies ergodicity.
C ONJECTURE 2.7. (Pugh and Shub) Stable accessibility is C r -dense among volume-pre-
serving partially hyperbolic diffeomorphisms, for r > 1.
In [BW10], Burns and Wilkinson proved Conjecture 2.6 under the additional condition
of center bunching (which means that the hyperbolicity of the E s ⊕ E u -bundle is stronger
than the non-conformality of E c ). Previously, Dolgopyat and Wilkinson [DW03] had
proved that stable accessibility is C 1 -dense. Recently, Avila, Crovisier and Wilkinson have
shown that stable ergodicity is C 1 -dense among C r partially hyperbolic diffeomorphisms,
with r > 1.
T HEOREM 2.8. (Avila, Crovisier and Wilkinson [ACW14]) Stable ergodicity is C 1 -dense
among C r volume-preserving partially hyperbolic diffeomorphisms, r > 1.
In their work, they build on an approach by [HHTU11] in which this result was proved
for the case of two-dimensional center bundles. In [HHTU11], the Pesin homoclinic
classes were introduced, which were shown to be hyperbolic ergodic components of m;
these classes were made into one by the use of blenders. The use of this technique
for center bundles of higher dimensions is a non-trivial fact, which was overcome by
[ACW14] by using a different kind of blender, introduced by Moreira and Silva [MS12].
Other important advance was made by Burns–Dolgopyat–Pesin [BDP02], who proved a
version of 2.6: namely, that essential accessibility and positive center Lyapunov exponents
imply stable ergodicity.
In [HHU08b], Conjecture 2.2 was proved for one-dimensional center bundles. In
particular, this gives the following theorem.
T HEOREM 2.9. (Hertz, Hertz and Ures [HHU08b]) Stable ergodicity is C ∞ -dense among
volume-preserving partially hyperbolic diffeomorphisms on 3-manifolds.
In summary, the vast majority of three-dimensional partially hyperbolic
diffeomorphisms are ergodic and, moreover, are stably ergodic. Hence one could ask the
following question: can we classify the non-ergodic partially hyperbolic diffeomorphisms?
And also: are there manifolds where all partially hyperbolic diffeomorphisms are
ergodic? Can we classify all the 3-manifolds admitting non-ergodic partially hyperbolic
diffeomorphisms?
Initial evidence in this direction was obtained in [HHU08a].
T HEOREM 2.10. (Hertz, Hertz and Ures) If N is a 3-nilmanifold other than T3 , then all
conservative partially hyperbolic diffeomorphisms are ergodic.
The proof of this theorem involves the study of accessibility classes defined above. It
follows from [BW10] and [HHU08b] that, for C 2 -partially hyperbolic diffeomorphisms
in 3-dimensional manifolds, accessibility implies ergodicity. This fact is very interesting
since it allows us to convert an ergodic problem into a geometric problem: the study of the
set of accessibility classes. In other words, if the system has only one accessibility class,
then it is ergodic. In §3, a better description of these sets in 3-manifolds is given.
While proving Theorem 2.10, we got the impression that the only obstruction to
ergodicity is the existence of a proper compact accessibility class.
C ONJECTURE 2.11. (Ergodic conjecture: Hertz, Hertz and Ures (2008)) If a conservative
partially hyperbolic diffeomorphism of a 3-manifold is non-ergodic, then there is a 2-torus
tangential to E s ⊕ E u .
The importance of this kind of hyperbolic sub-dynamics will become apparent later
on (see §6). As we will see, these tori seem to be ‘behind’ a lot of interesting behavior
F IGURE 2. (1) The 3-torus (2) The mapping torus of −Id (3) The mapping torus of a hyperbolic automorphism.
in partially hyperbolic dynamics. Moreover, not every orientable manifold can support
any such sub-dynamics. In order to describe the 3-manifolds admitting these 2-tori, let
us recall the concept of mapping torus. If N is a closed manifold and g : N → N is a
diffeomorphism, we define the mapping torus of g as the manifold obtained by identifying
in N × [0, 1] the points (x, 1) with (g(x), 0).
As examples of these, we can mention the 3-nilmanifolds, which can be seen as the
1 k

mapping tori of automorphisms of the form 0 1 , with k ∈ Z. The particular case when
k = 0 gives the 3-torus. Examples of solvmanifolds are the mapping tori of hyperbolic
automorphisms A : T2 −→ T2 .
T HEOREM 2.12. [HHU11] If a partially hyperbolic diffeomorphism f : M 3 → M 3 has

a 2-dimensional embedded torus T which is tangential either to E s ⊕ E u , E c ⊕ E u or
E c ⊕ E s , then the ambient manifold M 3 can only be one of the following possibilities:
(1) the 3-torus T3 ;
(2) the mapping torus of −id : T2 → T2 ; or
(3) the mapping tori of hyperbolic automorphisms on 2-tori.
See Figure 2.
In the three cases, it is possible to find partially hyperbolic dynamics with embedded
tori tangential to E s ⊕ E u or E c ⊕ E u (E c ⊕ E s is analogous). We refer the reader to §4
to see how a torus tangential to E c ⊕ E u can be built in any of these examples. Finding
a partially hyperbolic diffeomorphism with a torus tangential to E s ⊕ E u in cases (1) and
(3) is trivial; in case (2), it is enough to consider a hyperbolic automorphism on T2 , B,
and take the diffeomorphism f ([x, t]) = [Bx, t]. This is the desired partially hyperbolic
diffeomorphism.
If Conjecture 2.11 is true, there would be very few 3-manifolds supporting non-ergodic
partially hyperbolic dynamics. We state this explicitly as a weaker conjecture.
C ONJECTURE 2.13. (Weak ergodic conjecture: Hertz, Hertz and Ures (2008)) The
only orientable 3-manifolds that admit a non-ergodic conservative partially hyperbolic
diffeomorphism are:
(2) the mapping torus of −id : T2 → T2 ; and
The following problem is also open.

Problem 2.14. Let f be a conservative non-ergodic partially hyperbolic diffeomorphism

in any of the manifolds (1), (2), (3) stated in Theorem 2.12. Prove that there exists a torus
tangential to E s ⊕ E u .
In §3, we collect the advances towards proving this conjecture, to the best of our
knowledge, and a description of the set of accessibility classes.
2.2. Dynamical coherence. In partially hyperbolic dynamics, the strong bundles E s

and E u are always integrable: that is, there are invariant foliations W s and W u , the stable
and unstable foliations, that are tangential to each of the strong bundles (see, for instance,
[BP74, HPS77]). However, the center bundle, E c , may or may not be integrable. In fact,
as Wilkinson noticed in [Wil98], the Anosov example of A. Borel, mentioned by Smale
in [Sma67], when viewed as a partially hyperbolic diffeomorphism, has a non-integrable
center bundle. The Borel–Smale example is explained in detail in [BW08]. For the sake
of completeness, let us briefly mention what it is about.
2.2.1. A non-dynamically coherent example. The aforementioned example is, in fact,

a non-toral Anosov automorphism on a six-dimensional nilmanifold. Let G 1 and G 2
be copies of the three-dimensional simply connected non-abelian nilpotent Lie group.
Consider bases X i , Yi , Z i of the corresponding Lie algebras Gi , i = 1, 2, with the bracket
condition
[X 1 , Y1 ] = Z 1 , [X 2 , Y2 ] = Z 2 . (2.7)
Now consider a hyperbolic automorphism A ∈ S L(2, Z) and let λ > 1 be one of its
eigenvalues. Consider f such that the derivative acts over the Lie algebra as given by
X 1 7 → λX 1 , X 2 7 → λ−1 X 2 ,
Y1 7 → λ2 Y1 , Y2 7 → λ−2 Y2 ,
Z 1 7 → λ3 Z 1 , Z 2 7 → λ−3 Z 2 .
The next step (which we will not explain) is to find a lattice 0 of G = G 1 × G 2 such
that it is invariant, so that the whole construction yields a diffeomorphism over the six-
dimensional nilmanifold G/ 0 (the coset space).
This example is, as a matter of fact, an Anosov diffeomorphism, but we may also look
at it as a partially hyperbolic diffeomorphism such that E s is the space generated by Z 2 ,
E u is the space generated by Z 1 and the center bundle, E c , is the space generated by
X 1 , X 2 , Y1 and Y2 .
Note that X 1 , X 2 , Y1 and Y2 do not satisfy the Frobenius condition, due to (2.7), and
therefore E c is not integrable: that is, there is no invariant foliation tangential to E c .
After examining this example, we may ask: Is the lack of the Frobenius condition
the only reason for non-integrability of the center bundle? What about the case of one-
dimensional E c , where the Frobenius condition is always trivially satisfied? Notice that
E c is a priori only Hölder, so Picard’s Theorem does not necessarily hold. The question
of whether a partially hyperbolic diffeomorphism exists with a one-dimensional non-
integrable center bundle has remained open since the 1970s. Let us define a stronger
concept of integrability of the center bundle.
Definition 2.15. A partially hyperbolic diffeomorphism is dynamically coherent if the

following two conditions are satisfied.
(1) There is an invariant foliation tangential to the distribution E c ⊕ E u .
(2) There is an invariant foliation tangential to the distribution E s ⊕ E c .
In 2009, Hertz, Hertz and Ures [HHU15b] provided the first example of a partially
hyperbolic diffeomorphism with a non-integrable one-dimensional center bundle. It is
a non-dynamically coherent example in a 3-torus. Let us mention that this example
strongly contrasts with a result by Brin, Burago and Ivanov [BBI09], which proves that all
absolutely partially hyperbolic diffeomorphisms of the 3-torus are dynamically coherent.
Absolute partial hyperbolicity is more restrictive than partial hyperbolicity and requires the
bound in (1.6) to be uniform: namely, that there exist λ < 1 < µ such that, for all x ∈ M
and unit vectors v σ ∈ E xσ , σ = s, c, u,
kD f (x)v s k ≤ λ ≤ kD f (x)v c k ≤ µ ≤ kD f (x)v u k. (2.8)
In [HHU15a], it was shown that if there is an invariant foliation tangential to the

distribution E c ⊕ E u , then this foliation cannot contain a compact leaf, that is, a 2-torus
tangential to E c ⊕ E u (see Theorem 4.4). Note that, in principle, a cu-torus could coexist
with a cu-foliation, but cannot be part of it.
In the example mentioned below, in fact, the distribution E c ⊕ E u is uniquely integrable
at every point of T3 minus a 2-torus, which is tangential to E c ⊕ E u . This shows that no
invariant foliation exists which is tangential to E c ⊕ E u , and hence the example is non-
dynamically coherent. We refer the reader to Figure 7 and the caption below.
We will now prove the following theorem.
T HEOREM 2.16. [HHU15b] There exists a partially hyperbolic diffeomorphism f : T3 →
T3 such that:
(1) E c is not integrable;
(2) there is no invariant foliation tangential to the distribution E c ⊕ E u ; and
(3) there is an invariant two-dimensional torus T tangential to the distribution E c ⊕ E u .
Moreover, there is a C 1 -open neighborhood U of f such that all g in U satisfy conditions
(1) and (2).
Again, a two-dimensional periodic torus appears. How relevant is this object? We
conjecture that these tori might be the only obstructions to dynamical coherence.
C ONJECTURE 2.17. (Dynamical coherence conjecture: Hertz, Hertz and Ures (2009)) If
a partially hyperbolic diffeomorphism of a 3-manifold is not dynamically coherent, then
there is a 2-torus tangent to either E c ⊕ E u or E s ⊕ E c .
Again, if this conjecture were true, the only manifolds supporting non-dynamically
coherent dynamics would be the ones listed in Theorem 2.12. We state this explicitly as a
weaker conjecture.
C ONJECTURE 2.18. (Weak dynamical coherence conjecture: Hertz, Hertz and Ures
(2009)) The only orientable 3-manifolds supporting non-dynamically coherent dynamics
are:

(2) the mapping torus of −id : T2 → T2 ; and
Conjecture 2.17 was proved to be true in the 3-torus by Potrie in his PhD Thesis [Pot12].
This result was extended to manifolds with solvable fundamental group by Hammerlindl
and Potrie in [HP13].
T HEOREM 2.19. (Hammerlindl and Potrie [HP13]) Let f be a partially hyperbolic diffeo-
morphism of a 3-manifold with solvable fundamental group. If f is not dynamically
coherent, then there exists a 2-torus tangential to either E c ⊕ E u or E s ⊕ E c .
Hammerlindl and Potrie have hence proved that if the weak Conjecture 2.18 is true, then
the strong Conjecture 2.17 also holds. After Hammerlindl and Potrie, in order to prove
Conjecture 2.17, it would be enough to show that all partially hyperbolic diffeomorphisms
in 3-manifolds with non-solvable fundamental group are dynamically coherent.
Hammerlindl and Potrie’s result is the strongest and most complete one in this direction
up to this date.
In §4, we sketch a proof of this fact for the simplest case (see also Hammerlindl and
Potrie’s survey [HP15] for a more complete explanation on this and classification topics).
2.3. Classification. The third main topic in the study of partially hyperbolic dynamics
in 3-manifolds is their classification. As early as 2001, Enrique Pujals proposed the
following conjecture.
C ONJECTURE 2.20. (Classification conjecture: Pujals (2001); see [BW05a]) If a partially
hyperbolic diffeomorphism of a 3-manifold is transitive, then is (finitely covered by) one of
the following:
(1) a perturbation of the time-one map of an Anosov flow;
(2) a skew product; or
(3) a DA-diffeomorphism.
In 2009, Hammerlindl showed in his PhD Thesis [Ham13] that every absolutely
partially hyperbolic diffeomorphism of T3 is leafwise conjugate to its linearization. Let us
define this concept.
Definition 2.21. (Leaf conjugacy) Two dynamically coherent partially hyperbolic
diffeomorphisms f, g : M → M are leafwise conjugate if there exists a homeomorphism
h : M → M carrying center leaves of f to center leaves of g: that is, h(W cf (x)) =
Wgc (h(x)), such that
h( f (W cf (x))) = g(h(W cf (x))).
Note that, under the hypothesis of Hammerlindl’s thesis (absolute partial hyperbolicity
in T3 ), there is always dynamical coherence, due to a result by Brin, Burago and Ivanov
[BBI09].
Hammerlindl’s thesis suggested that perhaps dynamical coherence was a more suitable
hypothesis for a classification of three-dimensional partially hyperbolic dynamics. This,
together with our example [HHU15b], led us to propose the following conjecture.
C ONJECTURE 2.22. (Classification conjecture: Hertz, Hertz and Ures (2009)) Let f be a
partially hyperbolic diffeomorphism of a 3-manifold.
If f is dynamically coherent, then it is (finitely covered by) one of the following:
(1) a perturbation of a time-one map of an Anosov flow, in which case it is leafwise
conjugate to an Anosov flow;
(2) a skew product, in which case it is leafwise conjugate to a skew product with linear
base; or
(3) a DA, in which case it is leafwise conjugate to an Anosov diffeomorphism of T3 .
If f is not dynamically coherent, then there are a finite number of 2-tori tangential either to
E c ⊕ E u or to E s ⊕ E c , and the rest of the dynamics is trivial, as in the non-dynamically
coherent example [HHU15b] (see also §4).
Both conjectures are true in certain manifolds, as was proved by Hammerlindl and
Potrie [HP13].
T HEOREM 2.23. (Hammerlindl and Potrie [HP13]) Conjectures 2.20 and 2.22 are true
for partially hyperbolic diffeomorphisms in 3-manifolds with solvable fundamental group.
The classification conjectures have motivated much work. However, very recently,
Bonatti, Gogolev, Parwani and Potrie found both a dynamically coherent example and
a transitive example that are not leafwise conjugate to any of the above models, proving
both classification conjectures wrong [BPP14, BGP15] (see §5).
Question 2.24. Is it possible to classify partially hyperbolic dynamics in 3-manifolds,

modulo leaf conjugacies?
In the next sections, we shall develop more deeply the concepts mentioned above. We
have tried to make each section as self-contained as possible, which may mean that some
definitions have been repeated.
3. Ergodicity
To establish the ergodicity of partially hyperbolic maps, the most general method available
is the so-called Hopf method. To explain it, we first recall the following theorem.
T HEOREM 3.1. (Stable manifold theorem) Let M be a 3-manifold and let f ∈ Diffr (M) be
partially hyperbolic. Then there exist continuous foliations W s = {W s (x)}x∈M and W u =
{W u (x)}x∈M tangential to E s and E u , respectively, called the stable and the unstable
foliations. Their leaves are C r -immersed lines.
See [HPS77, Theorem 4.1]. We point out that, while their leaves are as smooth as f , the
foliations W s , W u are seldom differentiable [Ano67, p. 201]. Their transverse regularity
is only Hölder [PSW97], in general.
Using the Birkhoff ergodic theorem, it is not hard to see that a conservative
diffeomorphism f is ergodic if and only if, for every continuous observable ϕ : M → R,
the Birkhoff average
n−1
1X
ϕ̃(x) = lim ϕ ◦ f k (x) (3.9)
|n|→∞ n
k=0
is almost everywhere constant. But ϕ̃(x) coincides almost everywhere with

n−1
1X
ϕ + (x) = lim ϕ ◦ f k (x), (3.10)
n→∞ n
k=0
which is constant on stable manifolds (using uniform continuity of ϕ). Analogously, ϕ̃(x)
coincides almost everywhere with
n−1
1X
ϕ − (x) = lim ϕ ◦ f −k (x), (3.11)
n→∞ n
k=0
which is constant on unstable manifolds.
Suppose that f is not ergodic. Then there would be a continuous observable ϕ for
which ϕ̃ is not almost everywhere constant, and thus neither are ϕ + nor ϕ − . Hence there
exist two positive measure-invariant sets A and B and α ∈ R such that ϕ + (x) ≥ α for all
x ∈ A and ϕ − (x) < α for all x ∈ B. Note that A is saturated by stable manifolds while B
is saturated by unstable manifolds.
Assume that f is an Anosov diffeomorphism. Let x be a point of A such that almost
all points w ∈ W s (x) satisfy ϕ − (w) = ϕ + (w) = ϕ + (x). Such an x exists because the
stable foliation is absolutely continuous [Ano67] and ϕ + (w) = ϕ − (w) almost everywhere.
Consider y, a point belonging to the support of B, and V , a product neighborhood
containing y. On the one hand, observe that W s (x) intersects V and let WV be a connected
component of W s (x) ∩ V . On the other hand, m(V ∩ B) > 0, and then the absolute
continuity of the stable foliation implies that there is a local stable manifold T ⊂ V
such that m T (T ∩ B) > 0, where m T is the Lebesgue measure of T . If we call h u the
unstable holonomy in V sending T to WV , m WV (h u (T ∩ B)) > 0. Since B is u-saturated,
h u (T ∩ B) ⊂ B. We obtain a contradiction with the fact that almost every point in WV
satisfies ϕ − (w) = ϕ + (w). This is basically the Hopf argument.
In brief, there are two fundamental ingredients in the Hopf argument for an Anosov
map.
(1) There is a way of joining any pairs of points through a curve that is piecewise either
a stable or an unstable leaf.
(2) The stable and unstable foliations are absolutely continuous and completely
transversal.
3.1. Accessibility, a property that implies ergodicity. We would like to apply the
previous method to a general partially hyperbolic system, that is, when there is some non-
trivial center direction. To begin with, observe that, in general, it is not true that any two
pairs of points can be joined by a concatenation of stable and unstable leaves. For example,
if we consider the partially hyperbolic diffeomorphism 1 1 × id in T2 × S1 , then any
2 1

path consisting of a concatenation of stable and unstable leaves would be contained in

a single 2-torus. We fix f : M → M C 2 to be conservative partially hyperbolic and
call c : [0, 1] → M an su-path if it is piecewise C 1 and, for every t, where defined,
c0 (t) ∈ E s ∪ E u .
Definition 3.2. For a point x ∈ M, its accessibility class is the set
AC(x) := {y : ∃c : [0, 1] → M su-path such that c(0) = x, c(1) = y}.
The map f is accessible if the partition by accessibility classes is trivial and is essentially
accessible if the partition by accessibility classes is ergodic (that is, any Borel set saturated
by accessibility classes has either volume zero or one).
In the example above, each invariant 2-torus is an accessibility class. When there is
only one accessibility class, we will say that f has the accessibility property. From now
on, let us suppose this is our case. As for (2), absolute continuity of the strong foliations is
also satisfied [PS72], but complete transversality is not (due to the presence of the center
direction).
This problem can be overcome if the holonomies are regular enough. For instance,
Sacksteder used accessibility and Lipschitzness of the stable and unstable holonomies to
prove ergodicity of linear partially hyperbolic automorphisms of nilmanifolds [Sac70].
More generally, Brin and Pesin proved that accessibility and Lipschitzness of the stable and
unstable foliations imply ergodicity (in fact, Kolmogorov), in the following way [BP74,
Theorem 5.2, p. 204]; see also [GPS94]. If A and B are defined as before, consider a
density point x in A and a density point y in B. Take an su-path joining x and y. Call h
a global holonomy map from x to y: that is, h is a local homeomorphism that takes points
in a neighborhood U of x, slides them first along a stable segment, then along an unstable,
then along a stable again, etc. until reaching a neighborhood V of y, where all the su-
paths are near the original su-path joining x and y. Since A is essentially su-saturated,
h(A ∩ U ) = A ∩ V modulo a zero set. Since h can be chosen to be Lipschitz, there exists
a constant C > 1 such that, for each measurable set E ⊂ U , and for each sufficiently small
r > 0,
1
m(E) < m(h(E)) < Cm(E), (3.12)
C
Br/C (y) ⊂ h(Br (x)) ⊂ BCr (y). (3.13)
This implies that

m(BCr (y) ∩ A) m(h(Br (x) ∩ A)) 1 m(Br (x) ∩ A)
≥ ≥ →∞
m(BCr (y) \ A) m(h(BC 2 r (x) \ A)) C 2 · C 0 m(Br (x) \ A)
since m(BC 2 r (x) \ A) ≤ C 0 m(Br (x) \ A) for some positive constant C 0 . From this we get
that y is also a density point of A. This is absurd, since y was a density point of B,
complementary to A modulo a zero set.
This is essentially how the Hopf argument would work in the partially hyperbolic
setting. However, Lipschitzness of the holonomy maps is a very strong hypothesis, not
satisfied for most of the partially hyperbolic diffeomorphisms.
The idea of Grayson, Pugh and Shub [GPS94], and later improved by [Wil98, PS00,
HHU08b, BW10], is to show that the stable and unstable holonomies do preserve density
points according to another basis, different from intervals, called juliennes. These sets
are dynamically defined and constitute Vitali bases. Burns and Wilkinson made an
improvement of this ergodicity argument in [BW05b]. We will roughly sketch it. Consider
for a point x a small center segment, and saturate by local unstable leaves; to gain better
control of the size of these unstable segments we pre-iterate n times the local unstable
manifold of f n (x) (of a convenient size). The resulting set is then saturated by locally
stable manifolds. This small prism is called s-julienne and is denoted by Jnsuc (x). The
subscript n essentially tells the size of the julienne and, in particular, everything is chosen
so that m(Jnsuc (x)) −−−→ 0. An s-julienne density point of a set E is a point x such that
n7→∞
m(Jnsuc (x) ∩ E)
lim = 1. (3.14)
n→∞ m(Jnsuc (x))
The scheme is to consider the sets A and B that we considered above, and prove that:
(1) the s-julienne density points of A (and of any essentially u-saturated set) coincide
with the Lebesgue density points of A; and
(2) the s-julienne density points of A (and of any essentially s-saturated set) are
preserved by stable holonomies.
An analogous statement is proved for A with respect to u-julienne density points, which
are defined with respect to the local basis obtained by locally saturating a small center
segment first in a dynamic way by stable leaves, and then by unstable leaves. As the stable
and unstable holonomies preserve the Lebesgue density points of A, if the diffeomorphism
has the accessibility property, then A is equal to M modulo a zero set. This proves that the
system is ergodic.
T HEOREM 3.3. [BW10, HHU08b] If f ∈ Diff2m (M 3 ) is partially hyperbolic and satisfies

the accessibility property, then it is ergodic.
In fact, Burns and Wilkinson proved a much more general result.
T HEOREM 3.4. [BW10] If f is a partially hyperbolic diffeomorphism (with any center

dimension) satisfying the accessibility property and the center bunching property, then f
is ergodic.
A diffeomorphism is said to satisfy the center bunching property if

m(D f (x)|E c ) kD f (x)|E c k
kD f (x)|E s k < ≤ < m(D f (x)|E u ), (3.15)
kD f (x)|E c k m(D f (x)|E c )
where m(T ) = kT −1 k−1 .

Hence Theorem 3.4 implies Theorem 3.3. Also, when the center bundle is one-
dimensional, it is always locally integrable to center curves, so the fake foliations are
not necessary to build local juliennes. Hertz, Hertz and Ures, in [HHU08b], show an
alternative way of proving this result in this particular case.
There are two innovations in [BW10]. One is the argument outlined above Theorem 3.3,
which uses a much weaker center bunching than in [PS00] paper. This innovation had
already been published in [BW05b], and was essential in the proof of Theorem 3.3 in
[HHU08b]. In [BW05b], however, dynamical coherence was still a hypothesis, and the
task in [HHU08b, Theorem 3.3] consisted of removing this for the center dimension one
case. The second innovation in [BW10] was precisely to remove the dynamical coherence
hypothesis for any center dimension. This was accomplished by means of ‘fake foliations’,
which is a very delicate technical tool.
F IGURE 3. An su-path from y to z
3.2. Properties of accessibility classes. We want to describe precisely non-ergodic

partially hyperbolic diffeomorphisms, and it is possible that this only occurs when there is
a compact accessibility class (see Conjecture 2.11): that is, when there is a torus tangential
to E s ⊕ E u (in fact, we conjecture that there must be at least two such tori).
Since accessibility implies ergodicity, in order to describe non-ergodic partially
hyperbolic diffeomorphisms, it seems reasonable to look at the non-accessible ones. And,
even more precisely, we will study the structure of the set of non-open accessibility classes.
T HEOREM 3.5. [HHU08b] For each x ∈ M 3 , its accessibility class AC(x) is either an
open set or an immersed surface. Moreover, 0( f ), the set of non-open accessibility classes
of f , is a compact codimension-one laminated set whose laminae are the accessibility
classes.
Remark 1. This theorem still holds for partially hyperbolic diffeomorphisms with center
dimension one.
Let us begin with a local description of open accessibility classes.
P ROPOSITION 3.6. For any point x ∈ M, the following statements are equivalent.
(1) AC(x) is open.
(2) AC(x) has non-empty interior.
(3) AC(x) ∩ Wlocc (x) has non-empty interior for any choice of W c (x).
loc
Proof. (2) ⇒ (1) Let y be in the interior of AC(x) and consider any point z in AC(x).
Then there is an su-path from y to z with points y = x0 , x1 , . . . , x N = z such that xn
and xn+1 are either in the same s-leaf or in the same u-leaf (see Figure 3). Let U be
a neighborhood of y contained in AC(x) and suppose that, for instance, y = x0 and x1
belong to the same s-leaf. Then U1 = W s (U ) is an open set contained in AC(x) which
contains x1 , so x1 is in the interior of AC(x). Indeed, W s is a C 0 -foliation, so the s-
saturation of an open set is open.
Now x1 and x2 belong to the same u-leaf. If we consider U2 = W u (U1 ), then U2 is an
open set contained in AC(x) and containing x2 in its interior. Defining inductively Un as
W s (Un−1 ) or W u (Un−1 ) according to whether xn belongs to the s- or the u-leaf of xn−1 ,
we obtain that all xn belong to the interior of AC(x); in particular, z. This proves that
AC(x) is open.
(1) ⇒ (3) Follows directly from the definition of relative topology.
(3) ⇒ (2) Let V be an open set in AC(x) ∩ Wlocc (x), relative to the topology of W c (x).
loc
Then W (V ) is contained in AC(x) and contains a disc D sc of dimension s + c transverse
s
F IGURE 4. An open accessibility class.
F IGURE 5. An accessibility class in 0( f ).
to E u . This implies that W u (D sc ) is contained in AC(x) and contains an open set.

Therefore AC(x) has non-empty interior.
Let O( f ) be the set of open accessibility classes, which is, obviously, an open set. Then
its complement, 0( f ), is a compact set. Let us see that it is laminated by the accessibility
classes of its points.
For any point x ∈ M, consider a local center leaf Wloc c (x). Locally saturate it by stable
leaves: that is, take the local stable manifolds of all points y ∈ Wlocc (x), to obtain a small
(s + c)-disc Wloc sc (x). Now locally saturate W sc (x) by unstable leaves to obtain a small
loc
neighborhood Wloc usc (x) (see Figure 4).
usc (x), consider the map
On Wloc
usc
pus : Wloc (x) → Wloc
c
(x), (3.16)
which is defined in the following way. Given y ∈ Wloc usc (x), there exists a unique point
pu (y) in the disc W sc (x) that belongs to the local unstable manifold of y. Since Wloc sc (x) is
the local stable saturation of Wlocc (x), then p (y) ∈ W sc (x) is in the local stable manifold
u loc
of a unique point p (y) in Wloc (x): that is, pus (y) is the point obtained by first projecting
us c
along unstable manifolds onto Wloc sc (x) and then projecting along stable manifolds onto
Wloc (x). Since the local stable and unstable foliations are continuous, psu is continuous.
c
Let AC x (y) be the connected component of AC(y) ∩ Wloc usc (x) that contains y. The
points of AC x (y) are the points that can be accessed by su-paths from y without getting
out from Wlocusc (x) (see Figures 4 and 5). Then we have the following local description of
accessibility classes of points in 0( f ).

c (x) such that y ∈ 0( f ), AC (y) = p −1 (y).
L EMMA 3.7. For any y ∈ Wloc x su
Proof. Let y be a point in Wloc c (x). Then p −1 (y) = W u (W s (y)), which is clearly
su loc loc
contained in AC x (y). But also, psu (AC x (y)) = y. Indeed, if psu (z) were different from y,
for some z ∈ AC x (y), we would have a situation as described in Figure 4, because, since
psu is continuous and AC x (y) is connected, psu (AC x (y)) is connected. If psu (AC x (y))
contained another point, then it would contain a segment, which has non-empty interior
c (x). Proposition 3.6 then would imply that AC(y) is open, which is absurd, since
in Wloc
y ∈ 0( f ). This proves that also AC x (y) is contained in psu
−1 (y).
Hence, due to Lemma 3.7 above, for each x ∈ 0( f ),
AC x (x) = psu
−1
(x) = Wloc
u
(Wloc
s
(x)) ≈ Wloc
u
(x) × Wloc
s
(x).
s (x) and W u (x) are (evenly sized) embedded segments that vary continuously with
Wloc loc
respect to x ∈ M (see Hirsch, Pugh, Shub [HPS77, Chapters 4 and 5]). This implies that
0( f ) 3 x 7 → AC x (x) is a continuous map that assigns to each x an evenly sized 2-disc. To
finish the description of accessibility classes, let us introduce the following definition.
Definition 3.8. The foliations W s and W u are jointly integrable at a point x ∈ M if there
exists δ > 0 such that, for each z ∈ Wδs (x) and y ∈ Wδu (x),
u
Wloc (z) ∩ Wloc
s
(y) 6 = ∅.
See Figure 5 for an illustration of a point of joint integrability of W s and W u .
Then Lemma 3.7 and the discussion above imply the following proposition.
P ROPOSITION 3.9. A point x belongs to 0( f ) if and only if W s and W u are jointly

integrable at all points of AC(x).
Indeed, if x belongs to 0( f ), then, for all y ∈ AC(x) ⊂ 0( f ), psu (AC y (x)) = {y}. In
particular, if z ∈ Wδu (y) and w ∈ Wδs (y), then Wloc
s (z) ∩ W u (w) 6 = ∅. On the other hand,
loc
s u
if W and W are jointly integrable at all points of AC(x), then AC(x) is a lamina, due to
the explanation above (the coherence of the charts φx defined above depends only on the
joint integrability of W s and W u ). Moreover, this two-dimensional lamina is transverse to
c (x), and so AC(x) ∩ W c (x) cannot be open. Proposition 3.6 implies that AC(x) is
Wloc loc
not open, so x ∈ 0( f ).
The following lemma shows that, in fact, the laminae of 0( f ), that is, the accessibility
classes of points in 0( f ), are C 1 .
L EMMA 3.10. [Did03, Lemma 5] If W s and W u are jointly integrable at x, then the set
su
Wloc (x) = {W u (z) ∩ W s (y) : with z ∈ Wδs (x) and y ∈ Wδu (x)},
where δ > 0 is as in the definition of joint integrability (Definition 3.8), is a two-

dimensional C 1 -disc that is everywhere tangential to E s ⊕ E u .
In order to prove Lemma 3.10, we shall use the following result by Journé.
T HEOREM 3.11. [Jou88] Let F h and F v be two transverse foliations with uniformly
smooth leaves on an open set U . If η : U → M is uniformly C 1 along F h and F v , then η
is C 1 on U .
Proof of Lemma 3.10. Let D be a small, smooth two-dimensional disc containing x and
transverse to E xc . Consider a one-dimensional smooth foliation of a small neighborhood
N of x, transverse to D. If D is sufficiently small, there is a smooth map, π : N → D,
defined as projection along this smooth one-dimensional foliation. Note that Wloc su (x) can
be seen as the graph of a continuous function η : D → N .

We produce a grid on D in the following way. The horizontal lines are the projections
of the stable manifolds W s (y), with y ∈ Wδu (x): that is, the horizontal lines are of the
form π(Wloc s (η(v))), with v ∈ D. Analogously, the vertical lines are the projections of
the unstable manifolds Wloc u (z), with z ∈ W s (x): that is, the vertical lines are of the form
δ
π(Wlocu (η(w))), with w ∈ D.
Now v 7 → Wlocs (η(v)) and w 7 → W u (η(w)) are continuous in the C 1 -topology: that is,
loc
for close v, we obtain close Wloc s (η(v)) in the C 1 -topology (E s is a continuous bundle).
Since π is smooth, we also obtain that F h = {π(Wloc s (η(v)))}

v∈D , the horizontal partition
v
of D, and F = {Wloc (η(w))}w∈D , the vertical partition of D, are transverse foliations
u
continuous in the C 1 -topology.

But η is uniformly C 1 along F h , since η along a leaf F h (v0 ) = π(Wloc s (η(v )))
0
is exactly Wloc (η(v0 )). Indeed, η ◦ π : Wloc (x) → Wloc (x) is the identity map and
s su su
s (η(v )) is a smooth manifold. Analogously, we obtain that η is uniformly C 1 along
Wloc 0
v
F . Hence, by Theorem 3.11, η is C 1 .
3.3. Properties of the lamination 0( f ) of non-open accessibility classes. In order to

prove the ergodicity Conjecture 2.11, a possible strategy is to accurately describe the
lamination 0( f ). More precisely, one would like to see that non-accessibility implies
the existence of a compact (toral) accessibility class. However, the state of the art is given
by the following theorem.
T HEOREM 3.12. [HHU08a, Theorem 1.6] Let f : M 3 → M 3 be a conservative partially

hyperbolic diffeomorphism that is not accessible. Then one of the following possibilities
holds.
(1) There is a compact accessibility class (a torus tangential to E s ⊕ E u ).
(2) There exists an invariant sub-lamination 3 ⊂ 0( f ) of M that trivially extends to a
(not necessarily invariant) foliation without compact leaves. Moreover, if 3 6 = M,
the boundary leaves of 3 are periodic, have Anosov dynamics and periodic points
are dense in each boundary leaf with the intrinsic topology.
(3) 0( f ) is a minimal foliation.
With respect to item (2), a leaf L of a lamination 3 is a boundary leaf if there is

a transverse segment to L containing a subsegment α with an endpoint in L and such
that α ∩ 3 = ∅. In [HHU08a], it is proved that boundary leaves are periodic in the
conservative setting and, moreover, that periodic points are dense in each boundary leaf
with the intrinsic topology.
If Case (1) holds, then Conjecture 2.11 is true. We conjecture that Case (2) is
not possible; more precisely, we conjecture that each boundary leaf should be a torus.
Answering the following question in the affirmative would rule out Case (2).
Question 3.13. Let L be a complete immersed surface in a 3-manifold such that there is
an Anosov dynamics on L, where:
(1) each stable and unstable manifold is complete, and angles between stable and
unstable manifolds are bounded;
(2) periodic points are dense with the intrinsic topology; and
(3) the stable and unstable manifold of each periodic point are dense in L with the
intrinsic topology.
Is L the 2-torus?
Case (3) of Theorem 3.12 means that each leaf of 0( f ) is dense. We conjecture
that, in this case, in fact, f is essentially accessible, which means that each set that is
a union of accessibility classes has either measure one or zero. Essential accessibility in
dimension three implies ergodicity [BW10, HHU08b]. If this could be established, then
Conjecture 2.11 would be proved true. Finding a counterexample, however, would be very
interesting.
Question 3.14. Is there an example of a partially hyperbolic diffeomorphism in a 3-
manifold such that the accessibility classes of f form a minimal foliation, but f is not
essentially accessible?
Since in Case (1) of Theorem 3.12, Conjecture 2.11 follows trivially, we would like to
better describe what happens in Cases (2) and (3). The following describes the accessibility
classes in these cases.
T HEOREM 3.15. If f has no compact accessibility class, then the π1 of each accessibility
class injects in π1 (M).
Proof. The result follows almost directly from the following Theorem by Novikov.
T HEOREM 3.16. (Novikov) Let M be a compact orientable 3-manifold and let F be a
transversely orientable codimension-one foliation without Reeb components. Then, for
each leaf L in F, π1 (L) injects in π1 (M).
A Reeb component of a foliation is a solid torus subfoliated by planes, as in Figure 6.
Question 3.17. Does Theorem 3.15 hold without assuming that there are no compact
accessibility classes?
If 0( f ) = M, then we are already in the hypothesis of Theorem 3.16, since the fact
that 0( f ) has no compact leaves precludes the existence of Reeb components. The rest of
the theorem follows by proving that if 0( f ) 6 = M has no compact leaves, then it can be
extended to a foliation without Reeb components. This follows almost immediately from
[HHU08a, Theorem 4.1].
F IGURE 6. Reeb component.
T HEOREM 3.18. (Hertz, Hertz and Ures) If 3 ⊂ 0( f ) is an orientable and transversely

orientable f -invariant sub-lamination without compact leaves such that 3 6 = M, then all
closed complementary regions of 3 are I -bundles.
By taking finite coverings, we may assume that 0( f ) itself is orientable and transversely
orientable. 0( f ) 6 = M has no compact leaves, therefore its complementary regions are I -
bundles. This allows us to extend 0( f ) to a foliation in a trivial way, by ‘copying’ the
boundary leaves. This means that each complementary region is of the form L × [0, 1],
where L is a boundary leaf of 0( f ). We foliate each complementary region by considering
leaves of the form L × {t}, with t ∈ [0, 1].
4. Dynamical coherence
It turns out that ergodicity in our setting is tightly related to integrability of the invariant
bundles. As we explained before, the bundles E s , E u are always integrable. The
integrability of the center bundle E c , on the other hand, cannot be always guaranteed, even
in our setting. This was a long-standing problem and was recently solved in [HHU15b]
(see Theorem 2.16, rewritten below).
Let us recall dynamical coherence, which was stated in Definition 2.15. A partially
hyperbolic diffeomorphism is dynamically coherent if there is an f -invariant foliation
tangential to E s ⊕ E c (the center-stable foliation) and an f -invariant foliation tangential
to E c ⊕ E u (the center-unstable foliation). Note that, in this case, the center bundle is also
integrable: an f -invariant foliation tangential to E c is obtained by simply intersecting the
center-stable and center-unstable leaves and taking connected components. This is called
the center foliation.
P ROPOSITION 4.1. If f : M 3 → M 3 is a partially hyperbolic diffeomorphism whose

center bundle is C 1 , then f is dynamically coherent.
Proof. First, observe that W c is f -invariant: if c : [0, 1] → W c (x) is a differentiable

curve with c(0) = x, then f ◦ c : [0, 1] → M is a differentiable curve tangential to E c ,
and hence, by uniqueness of solutions of differential equations, f ◦ c([0, 1]) ⊂ W c ( f ◦
c(0)) = W c ( f (x)).
Theorem 6.1 and Theorem 7.6 in [HPS77] imply that, through each leaf L of W c , there
exist immersed submanifolds W s (L) and W u (L) tangential to E cs and E cu , respectively,
saturated by the corresponding strong foliations. Again using uniqueness of solutions

of differential equations, one proves that the families W cs = {W s (L)} L∈W c , W cu =
{W u (L)} L∈W c are pairwise disjoint and, since their tangent spaces vary continuously, they
form foliations. Invariance follows, since W c , W s , W u are invariant.
More details about this can be found in [BW08]. When the center bundle is not
differentiable, we still have curves tangential to it as a consequence of Peano’s theorem.
This family of curves, however, is not assembled as a foliation, but it still can contain
relevant information (see [HHU15a] and [HHU08c]).
Problem 4.2. Find an example of a dynamically coherent partially hyperbolic

diffeomorphism that is not leafwise conjugate to a C 1 dynamically coherent one. Can
the examples in [BPP14] be adjusted to get one?
Another condition that guarantees dynamical coherence in T3 is absolute partial

hyperbolicity, a notion stronger than partial hyperbolicity, which was described in
Equation (2.8).
T HEOREM 4.3. (Brin–Burago–Ivanov [BBI09]) If f : T3 → T3 is absolutely partially

hyperbolic, then f is dynamically coherent.
However, this is not the general case, as we explain in the next subsection.
4.1. A non-dynamical coherent example.
T HEOREM 2.13. [HHU15b] There exists a partially hyperbolic diffeomorphism f : T3 →

T3 such that:
(1) there is no invariant foliation tangential to the distribution E c ⊕ E u ; and
(2) there is an invariant two-dimensional torus T tangential to the distribution E c ⊕ E u .
Moreover, there is a C 1 -open neighborhood U of f in Diff1 (M) such that all g in U satisfy
conditions (1) and (2).
Sketch. Let A : T2 → T2 be a hyperbolic linear map with eigenvalues λ < 1 < 1/λ. Take
u, a unit eigenvector corresponding to the eigenvalue λ. Consider also a north pole–south
pole function f : T → T such that
f (0) = 0, f (1/2) = 1/2,
f 0 (1/2) = σ < λ < 1 < ν = f 0 (0) < 1/λ

and a differentiable function φ : T → R.
Now construct a perturbation F of the Axiom-A map A × f by ‘pushing’ in the stable
direction of A, namely,
F(x, θ ) = (Ax, f (θ )) + (φ(θ )es , 0), φ(1/2) = 0,
where es is a unit vector in the E s direction of A.

Note that the strong unstable direction of A × f is unaltered by this perturbation and,
in particular, the strong stable manifold of the perturbation exists and coincides with the
strong stable manifold of the unperturbed map. Observe that the unperturbed map is not
partially hyperbolic. Now we study the other invariant directions.
We are seeking invariant directions of the derivative of F,
d F(x,θ) (v, t) = (Av, f 0 (θ )t) + (φ 0 (θ )tes , 0).
An invariant direction (inside the es × T cylinder) will be generated by a vector field of

the form (a(θ )es , 1) for some function a, and hence we need to solve
a( f (θ )) f 0 (θ ) = λa(θ ) + φ 0 (θ ). (4.17)
We are thus led to find a solution of the cohomological equation
b ◦ f = λb + φ
(the solution of (4.17) is just a = b0 ). One then checks that the following two functions are
solutions, that is,
∞
1X n
η(θ ) = λ φ( f −n θ ), (4.18)
λ
1
∞
1 X −n
ζ (θ ) = − λ φ( f n θ ), (4.19)
λ
0
and that the previous assumptions imply that η ∈ C 1 (T \ {1/2}), ζ ∈ C 1 (T \ {0}).

Let us define
E c (θ ) = span(η0 (θ)es , 1) for θ 6 = 21
and
E s (θ ) = span(ζ 0 (θ )es , 1) for θ 6 = 0.
Back to the invariant directions, note that, for generic φ, η0 (θ ) gets bigger as θ
approaches 1/2, and thus, if we can choose φ so that
lim η0 (θ ) = ∞, (4.20)
θ→1/2
we will get continuity for E c by defining
E c (θ = 12 ) = span{(es , 0)} = E sA × 0.
Arguing similarly, we define
E s (θ = 0) = E sA × 0,
and we will get a continuous bundle provided that we prove that
lim ζ 0 (θ ) = ∞. (4.21)
θ →0
Assume for now that we have proved that these bundles are continuous. Now we want
to show that T T3 = E s ⊕ E c ⊕ E u or, equivalently, that the angle between E s and E c is
not zero. What we need to show is that η0 6 = ζ 0 for θ 6 = 0, 1/2. Note that for, θ = 0, 1/2,
F IGURE 7. The figure above shows a center-stable leaf. The center leaves are unique for θ 6 = 1/2 and become
tangential to the center leaf at θ = 1/2, making a peak. This precludes the existence of a center foliation.
the angle is not zero, and hence it is not zero in a neighborhood of these points. But, by
the cohomological equations,
(η0 − ζ 0 ) ◦ f = λ(η0 − ζ 0 )
and, using the form of the dynamics of f , we conclude that the sign of η0 − ζ 0 is constant
in (0, 1/2) and (1/2, 1) and is clearly non zero. The following lemma is proved in detail
in [HHU15b].
L EMMA 4.3. There exists φ : T → R such that:

(1) the limits in (4.20) and (4.21) hold; and
(2) η0 has opposite signs in (0, 1/2) and (1/2, 1).
In particular, F is partially hyperbolic (but NOT absolutely partially hyperbolic).

Finally we prove that it is not dynamically coherent. Observe that, since the bundles only
depend on θ we obtain the stable, unstable and center manifolds (provided that this last
one exists) by translation.
Consider the function h : T 3 → T 2 given by
h(x, θ ) = x − η(θ )es .
Then F ◦ h = h ◦ A T and h is clearly surjective. Hence it is a semiconjugacy. Note that

we have a parametrization l x (θ ) of h −1 (h(x, 0)) given by
l x (θ ) = (x, 0) + (η(θ )es , θ ),
and hence the family of curves {l x (θ )} is tangential to E c if θ 6 = 1/2. For θ 6 = 1/2 the
bundle E c is uniquely integrable and hence its invariant curves are precisely the l x (θ ).
But, for θ = 1/2, E c = E sA × {0}, and hence its tangent curves have to be horizontal. Now
we use that η0 have different signs on the intervals (0, 1/2) and (1/2, 1) to conclude that
this family is not a foliation near θ = 1/2 and hence that the bundle E c is not integrable
(see Figure 7).
The example previously constructed is, in fact, robust, meaning that in a neighborhood
of it there are no dynamically coherent partially hyperbolic diffeomorphisms, which is a
surprising fact. This is a consequence of the fact that the invariant torus corresponding
to θ = 1/2 is a cu-torus and, in particular, that it is a normally hyperbolic torus. The

results of [HPS77] imply that this torus persists under perturbations, meaning that any
small perturbation of f has a cu torus. On the other hand, outside a neighborhood of this
cu-torus, there is a unique center foliation, which is persistent under perturbations, due
also to [HPS77]. This foliation is extended by invariance to all the complements of the
cu-torus. Hence, if there were a foliation, it would contain a cu-torus. This is not possible,
due to the following result.
T HEOREM 4.4. (Hertz, Hertz and Ures [HHU15b]) Let f : M 3 → M 3 be a dynamically
coherent partially hyperbolic diffeomorphism. Then the center unstable foliation does not
have any closed leaf.
Question: Does the same result hold in any dimension?
4.2. Non-dynamical coherence conjecture—state of the art. It has been proved by A.

Hammerlindl and Potrie [HP13] that the dynamical coherence Conjecture 2.17 is true in
tori, solvmanifolds and nilmanifolds and their finite covers.
T HEOREM 4.5. (Hammerlindl and Potrie [HP13]) If f is a non-dynamical coherent
diffeomorphism in a 3-manifold with (virtually) solvable fundamental group, then there
exists a 2-torus, tangential either to E s ⊕ E c or E c ⊕ E u . In particular, any partially
hyperbolic diffeomorphism with ( f ) = M in these manifolds is dynamically coherent.
This is the sharpest result concerning Conjecture 2.17 so far. Let us give a brief sketch
of the ideas used to establish this theorem in the simpler case where M = T3 . This result
has been proved by Potrie in his thesis (see [Pot12]), and we shall follow his arguments.
Consider f : T3 → T3 partially hyperbolic, and, by passing to a finite covering, it is no
loss of generality to assume that the bundles E σf are oriented and, furthermore, that f
preserves their orientation.
We will rely heavily in the seminal papers of Brin, Burago and Ivanov [BI08, BBI04,
BBI09]. The starting point is that the action in homology f ∗ : H1 (T3 ) → H1 (T3 ) is
partially hyperbolic [BI08, Theorem 1.2.]: namely, if A = f ∗ ∈ S L(3, R) has eigenvalues
λ1 , λ2 , λ3 , then |λ1 | < 1 < |λ3 |. There are two cases.
(1) |λ2 | = 1; this is the skew-product case.
(2) |λ2 | 6 = 1; in this case f is a DA.
It suffices to show the existence of an f -invariant foliation tangential to E cs (the other
case being analogous). It turns out that there is a natural candidate for F cs .
T HEOREM 4.6. [BI08, Theorem 4.1] There exists a family B cs = {B cs (x)}x∈M such that:
(1) each B cs (x) is an immersed boundaryless surface of class C 1 tangential to E cs ;
(2) for every x, x 0 ∈ M, x 6 = x 0 , the surfaces B cs (x), B cs (x 0 ) do not cross topologically,
that is, B cs (x 0 ) cannot intersect two different connected components of B \ B cs (x),
for any neighborhood B of x; and
(3) B cs is f -invariant.
The family B cs is what is called a branched foliation, and its elements are called leaves.
To prove that B cs is a genuine foliation, it suffices to show that it is unbranched: namely,
that any x ∈ M is contained in exactly one leaf of B cs . This fact is a direct consequence of
[BW05a, Proposition 1.6] and the remark afterwards.
Another important fact is that E cs is almost integrable: that is, there exists a foliation
(not necessarily invariant) that is transverse to E u . This concept of almost integrability has
been coined by R. Potrie and has proven useful in this context. Almost integrability of E cs
for all three-dimensional partially hyperbolic diffeomorphisms with orientable bundles had
been established by Burago–Ivanov.
T HEOREM 4.7. ([BI08], Key lemma) For every > 0 sufficiently small there exists a true
(not necessarily invariant) foliation Tcs such that the angle ∠(T Tcs , E cs ) is less than
. Moreover, there exists a continuous surjective map h cs : M → M that is -close to the
identity and sends the leaves of Tcs into leaves of B cs .
Potrie then shows that, for sufficiently small , the lifted foliations Ffu , Tgcs 3
to R have
global product structure (that is, any two leaves F ∈ F fu and T ∈ Tg cs
) intersect exactly in
one point). This has the following consequence.
T HEOREM 4.8. [Pot12, Proposition 8.4] B cs is unbranched.
The proof of this theorem relies on two geometric facts.

Fact 1. F fu is quasi-isometric: that is, there exist a, b > 0 such that, for every x, y ∈
R , y ∈ F u (x), the length l(x, y) of the interval with endpoints x, y contained in F
3 f fu (x)
satisfies
l(x, y) ≤ a|x − y| + b.
Fact 2. There exists an A-invariant plane P and R > 0 such that every leaf of B cs is
contained in an R-neighborhood of a plane parallel to P, and either:
(1) the projection of P in T3 is dense and the R-neighborhood of every leaf of B cs
contains a plane parallel to P; or
(2) the projection of P in T3 is a 2-torus. In this case, there exists a 2-torus tangential to
E s ⊕ E c.
Note that if f is a DA, only (1) above can hold. Let us see how the proof goes.
Proof. A modification of [BBI04, Proposition 3.7] gives that a codimension-one foliation

in T3 either has a Reeb component or there exists R > 0 and an A-invariant plane P ⊂ R3
such that every leaf of the lifted foliation is in a R-neighborhood of P. The projection of a
plane in T3 is either dense or a 2-torus. In the former case, every leaf of the lifted foliation
is parallel to a fixed translate of the plane, while, in the later case, there is a leaf of the
foliation in T3 which is a torus (see [Pot12, Theorem 5.3 and Proposition 5.6]).
As Tcs is transverse to W u , it cannot have a Reeb component and thus we can consider
the plane P, as above. Note that, by 4.7, the leaves of B cs are -close to the leaves of T cs
,
g g
and hence the first part of the claim follows.
Now it suffices to observe that if Tcs has a torus leaf, by Theorem 4.7 B cs also has a
torus leaf. Therefore, there is a 2-torus tangent to E s ⊕ E c .
Once the above facts are established, the proof of the theorem is carried out by standard
arguments. To finish the proof of the conjecture, one has to analyze the skew product and
the DA case separately. In both cases, with the machinery developed, it is not too hard to
check that the absence of tori tangential to E s ⊕ E c implies the global product structure
referred above, and thus implies dynamical coherence.
When the manifold M is a solvmanifold, but not a nilmanifold, the proof of Theorem 4.5
becomes technically harder, although some of the guidelines presented are still valid. It
relies on more background on foliation theory (codimension-one foliations of compact
solvmanifolds are reasonably well understood). Solvmanifolds of this type are covered by
torus bundles over the circle (a reasonable geometric object), and then there is a canonical
model (isotopic to the identity) to compare the dynamics. For more details and a complete
proof, we refer the reader to [HP13].
We also recommend the excellent survey by Hammerlindl and Potrie [HP15].
5. Classification
For many years, the only known examples of partially hyperbolic diffeomorphisms in
3-manifolds were the ones listed in §1.1: namely, time-one maps of Anosov flows,
skew products, DA-diffeomorphisms, and their perturbations. As stated in §2.3, this led
E. Pujals, in 2001, to conjecture that, for transitive ones, this was the complete list of
partially hyperbolic diffeomorphisms.
C ONJECTURE 2.20. (Pujals (2001)) Any transitive partially hyperbolic diffeomorphism in

a 3-manifold is finitely covered by a map which is conjugated to:
(1) a perturbation of the time-one map of an Anosov flow;
(2) a perturbation of a skew product; or
(3) a DA.
Two different particular cases of this conjecture were verified in the transitive setting by
Bonatti and Wilkinson [BW05a].
T HEOREM 5.1. (Bonatti–Wilkinson) Let f : M → M be a transitive partially hyperbolic

diffeomorphism.
(1) Assume that there exists some embedded circle c such that† f (c) = c, with the
property that, for some > o, the set
[ [
Ws (x) ∩ Wu (y)\c
x∈c y∈c
contains a connected component that is a circle. Then f is dynamically coherent and

finitely covered by a map which is conjugated to a circle extension of an Anosov map
(a topological skew product).
(2) Assume that f is dynamically coherent and that, for some > 0, each end of a center
leaf contained in [
Ws (x)
x∈c
is periodic. Then, the center foliation is fixed under f n and it supports a continuous
flow conjugate to an expansive transitive flow.
† Interestingly enough, the proof does not extend to the case where c is merely periodic.
It would be interesting to know if, in case (2), one can, in fact, take an Anosov flow and
thus settle Pujals’s conjecture for that case. This still remains an open problem.
More recently, a new type of non-dynamically coherent example was presented, which
is the one described in §4.1. The examples in §4.1 suggested another possibility.
C ONJECTURE 2.22. (Classification conjecture: Hertz, Hertz and Ures (2009)) Let f be a
partially hyperbolic diffeomorphism of a 3-manifold.
If f is dynamically coherent, then it is (finitely covered by) one of the following:
(1) a perturbation of a time-one map of an Anosov flow, in which case it is leafwise
conjugate to an Anosov flow;
(2) a skew product, in which case it is leafwise conjugate to a skew product with linear
base; or
(3) a DA, in which case it is leafwise conjugate to an Anosov diffeomorphism of T3 .
If f is not dynamically coherent, then there are a finite number of 2-tori tangential either to
E c ⊕ E u or to E s ⊕ E c , and the rest of the dynamics are trivial, as in the non-dynamically
coherent example [HHU15b].
Both conjectures have been proved false very recently by Bonatti, Gogolev, Parwani
and Potrie (see [BPP14, BGP15] and §5.1 for a description of these examples). However,
both conjectures are true in certain manifolds, as was proved by Hammerlindl and Potrie.
T HEOREM 5.2. (Hammerlindl and Potrie [HP13]) Both Conjectures 2.20 and 2.22 are
true on 3-manifolds with (virtually) solvable fundamental group.
Theorem 5.2 was first proved in tori by Hammerlindl in his thesis [Ham13], and it was
later extended to 3-manifolds with (virtualy) nilpotent groups by Hammerlindl and Potrie
in [HP14]. Finally, it was extended to 3-manifolds with (virtually) solvable groups by the
same authors in [HP13]. This is still in press.
In [HP13], it is proved that, for solvmanifolds, as stated in Conjecture 2.17, the
absence of tori tangential to either E s ⊕ E c or E c ⊕ E u implies dynamical coherence.
Observe that the existence of such a torus implies the existence of either a repelling or an
attracting periodic torus (see more details in §6). Transitivity precludes this possibility.
Therefore, for solvmanifolds, we can assume that there is dynamical coherence in both
Conjectures 2.20 and 2.22.
Let us give a flavor of how Theorem 5.2 is proved in the case of solvmanifolds with
non-virtually nilpotent fundamental group.
T HEOREM 5.3. [HP13] If f : M → M is a dynamically coherent partially hyperbolic

diffeomorphism on a solvmanifold whose fundamental group is not virtually nilpotent,
then a finite cover of a finite iterate of f is leafwise conjugate to the time-one map of a
suspension Anosov flow.
See Definition 2.21 for the definition of leaf conjugacy.

To start the proof of Theorem 5.3, one shows that any such manifold is finitely
covered by the mapping torus M A of a hyperbolic automorphism on T2 : that is M A =
T2 × R/ ∼ such that (Ax, t) ∼ (x, t + 1), where A is a hyperbolic automorphism of
T2 . Any diffeomorphism of M A has a finite iterate that is homotopic to the identity. This
is not hard to prove.
Now, on the universal cover of M A , there are model foliations Acs and Acu ; (v1 , t1 )
and (v2 , t2 ) belong to the same leaf of the foliation Acs if and only if v1 − v2 is in the
stable eigenspace of the automorphism A. Similarly, (v1 , t1 ) and (v2 , t2 ) belong to the
same leaf of the foliation Acu if and only if v1 − v2 is in the unstable eigenspace of the
automorphism A. In [HP13], it is seen that the lift to the universal cover of any foliation
without compact leaves is almost parallel to either Acs or Acu . Two foliations F and F 0
are almost parallel if there is a uniform bound R > 0 such that:
(1) for each leaf L ∈ F, there is a leaf L 0 ∈ F 0 such that d H (L , L 0 ) < R; and
(2) for each leaf L 0 ∈ F 0 , there is a leaf L ∈ F such that d H (L , L 0 ) < R,
where d H is the Hausdorff distance: that is
d H (L , L 0 ) = max sup d(x, L 0 ), sup d(y, L) .

x∈L y∈L 0
Now neither F cs nor F cu contain compact leaves [HHU15a], and therefore each one is
almost parallel to either Acs or Acu . They proceed then to show that if F cs is almost
parallel to Acs , then F cu is almost parallel to Acu . This step is more delicate.
Note that the center leaves of the model foliation (that is, the leaves in Ac that are
intersections of leaves Asc and Acu ) correspond to trajectories of an Anosov flow, which
is infinitely expansive. Infinite expansivity means that, for any two different points x and
y in the universal cover and any K > 0, there will be a time t ∈ R such that X t (x) and
X t (y) are K -apart. Therefore, any two such leaves Ac1 and Ac2 are at an infinite Hausdorff
distance. This implies that the almost-parallel relation defined above assigns to each center
leaf F c in the intersection of F sc and F cu a unique center leaf in Ac . Less trivially, there
is a unique leaf in F sc at a finite Hausdorff distance from each leaf in Asc and a unique
leaf in F cu at a finite distance from each leaf in Acu [HP13, Lemma 5.3]. Therefore, any
two center leaves F1c and F2c that are at a finite Hausdorff distance from each other must
be in the intersection of a single leaf of F sc and a single leaf of F cu .
Now let us assume that the center bundle E cf is orientable, for, otherwise, we can take
a finite cover. Then there exists a field X c tangential to E c , defining a flow ϕ on M A . We
claim that ϕ is an expansive flow.
Indeed, any two ϕ-trajectories that are at most ε-apart correspond to two center leaves
that are at a finite Hausdorff distance; hence, they are in the intersection of a single leaf of
F sc and a single leaf of F cu . This implies either that a stable leaf intersects (at least) twice
a center unstable leaf of F cu or that an unstable leaf intersects (at least) twice a center-
stable leaf of F sc . A classical argument, according to Novikov, implies the existence of a
compact leaf either in F cu or in F sc , which is a situation precluded by [HHU15a].
Finally, Brunella establishes, in [Bru93], that any expansive flow on a torus bundle is
leafwise conjugate to a transitive Anosov suspension, which concludes the classification
theorem in solvmanifolds.
Again, we refer the reader to the survey of Hammerlindl and Potrie, in order to gain a
deeper understanding of this topic.
5.1. Anomalous partially hyperbolic diffeomorphisms. Very recently, Bonatti,

Gogolev, Parwani and Potrie found both a dynamically coherent example and a transitive
example that are not leafwise conjugate to any of the above models, which proved both
conjectures wrong [BPP14, BGP15].
The idea behind both constructions is to perform a large ‘perturbation’ by composing
the time-one map of certain Anosov flows in an appropriate neighborhood with a map
of the form (g, Dg), where g is a Dehn twist and Dg is the derivative acting on the
unitary tangent bundle. This neighborhood has to be large enough so that the effect of
the derivative of the Dehn twist is negligible. We will explain this in more detail for the
case of the perturbations of the time-one map of the geodesic flow of surfaces of constant
negative curvature in Theorem 5.5.
The first family of (non-transitive) examples is obtained by modifying the Franks–
Williams’ construction [FW80b] of a non-transitive Anosov flow.
T HEOREM 5.4. (Bonatti, Parwani and Potrie [BPP14]) There is a closed orientable 3-
manifold M endowed with a non-transitive Anosov flow X t and a diffeomorphism f : M →
M such that:
• f is absolutely partially hyperbolic;
• f is robustly dynamically coherent;
• the restriction of f to its chain recurrent set coincides with the time-one map of the
Anosov flow X t ; and
• for any n > 0, f n is not isotopic to the identity.
The transitive examples are built on time-one maps of two different Anosov flows: the
Bonatti-Langevin example [BL94] and the geodesic flow of surfaces of negative constant
curvature (a similar construction works for the Handel–Thurston Anosov flow [HT80]).
After the statement of the theorem, we will give a brief outline of the construction for the
case of geodesic flows.
T HEOREM 5.5. (Bonatti, Gogolev and Potrie [BGP15]) There exist a closed orientable
3-manifold M and an absolutely partially hyperbolic diffeomorphism f : M → M that
satisfy that:
• M admits an Anosov flow;
• f n is not homotopic to the identity map for all n > 0;
• f is volume preserving; and
• f is robustly transitive and stably ergodic.
5.2. Unit tangent bundle of surfaces of negative curvature.. Here we give the main
ideas of the construction of the examples for this case. Let S be a surface and let g be
a Riemannian metric of curvature −1. Fix a simple closed geodesic γ . It is possible to
deform the hyperbolic metric in such a way the length of γ goes to zero. Indeed, there is
a sequence gn of metrics of curvature −1 such that lengthgn γ → 0. Then there are collars
of uniform length (for the metric gn ) of γ ; call them Cn . These collars become thinner
and thinner as n goes to infinity. This implies that Dehn twists ρn on these collars are very
close to isometries, for n large enough.
Now consider h n = Dρn ◦ ϕn , where ϕn is the time-one map of the geodesic flow of
gn and Dρn is the projectivization of the derivative of ρ. Since the partially hyperbolic
structure does not change with the metric (indeed, the partially hyperbolic structure
depends on the metric and, for all n, the metric on the universal cover is the same) this
will imply that h n is partially hyperbolic for n large enough. By constructing ρn with
some care, Dρn can be made volume preserving. Known results and techniques imply that
there is a stably ergodic and robustly transitive perturbation of h n .
Bonatti, Hammerlindl, Gogolev and Potrie have announced a generalization of the latter
construction (see [BGP15]). There is a natural homomorphism of the mapping class group
of a surface of genus greater than one, S, into the mapping class group of its unit tangent
bundle given by the projectivization I : MCG(S) → MCG(T1 S).
T HEOREM 5.6. Each mapping class of the image of I admits a volume-preserving
partially hyperbolic representative.
There are some open questions about the examples given by Theorem 5.5. The most
important is whether they are dynamically coherent.
Many new questions arise regarding the classification. Some of them are the given
below.
Question 5.7. Suppose that the fundamental group of M is not (virtually) solvable. If M
admits a partially hyperbolic diffeomorphism, does it support an Anosov flow?
Hammerlindl, Potrie and Shannon have announced that the answer is positive for Seifert
manifolds having fundamental group with exponential growth.
Question 5.8. Given a partially hyperbolic diffeomorphism on M, is there some
sort of inverse process of the previous construction leading to a partially hyperbolic
diffeomorphism isotopic to the identity? Or to a partially hyperbolic diffeomorphism leaf
conjugate (up to finite cover and iterate) to the time-one map of an Anosov flow?
6. A tool to better understand some partially hyperbolic dynamics: Anosov tori

Both the ergodicity and the integrability conjectures propose that the existence of a map
with some specific dynamical property leads to very rigid restrictions in the topology of
the ambient manifold. The reader may wonder why this is case and how one can attempt
to prove such types of results. We discuss these issues in this section.
The unifying link is, surprisingly, the existence of certain tori embedded in the manifold.
Definition 6.1. We say that the manifold M admits an Anosov torus if there exist a C 1 -
embedded torus T ⊂ M and a diffeomorphism φ : M → M such that:
(1) φ(T ) = T ; and
(2) φ|T is a linear hyperbolic automorphism†.
As we shall see below, not every manifold admits an Anosov torus.
Recall that a three manifold is irreducible if every embedded two-sphere bounds a three-
ball. We then have the following topological result.
† This is equivalent to the existence of φ such that φ|T is isotopic to an Anosov diffeomorphism, which holds if
and only if the action on the first homology group of the torus H1 (T ) is hyperbolic.
T HEOREM 6.2. (Hertz, Hertz and Ures [HHU11]) Assume that M is a compact,
irreducible 3-manifold supporting an Anosov torus. Then M is homeomorphic to:
(1) a 3-torus;
(2) the mapping torus of −Id : T2 → T2 ; or
(3) the mapping torus of an hyperbolic automorphism of T2 .
We remark that partial hyperbolicity is not required in Theorem 6.2. This theorem just
shows that there are actually very few 3-manifolds admitting Anosov tori. Now we apply
this result to the partially hyperbolic context. First, we note that irreducibility comes for
free in this setting.
L EMMA 6.3. If a 3-manifold M supports a partially hyperbolic diffeomorphism, then M

is irreducible.
Proof. A 3-manifold admitting a partially hyperbolic diffeomorphism has a codimension-

one foliation having neither Reeb components nor spherical leaves [BBI04]. This proves
the claim, since Rosenberg shows, in [Ros68], that any codimension-one foliation in
a reducible 3-manifold must have a Reeb component or a spherical leaf (see also
[Rou71]).
The following proposition describes Anosov tori that arise naturally in partially
hyperbolic dynamics (see [HHU15a] for more details).
P ROPOSITION 6.4. Let f : M → M be a partially hyperbolic diffeomorphism and assume

that there exists an f -invariant embedded torus T tangential to either E c ⊕ E u , E s ⊕ E c
or E u ⊕ E s . Then T is an Anosov torus.
Proof. Let g = f |T . In each of the different cases, g or g −1 preserves an expanding

foliation by lines, so, with no loss of generality, we will assume that kdg|E u k > 1. It
suffices to prove that g∗ : π1 (T ) ≈ Z2 → Z2 is hyperbolic.
By taking g 2 , if necessary, we can suppose that g preserves the orientation of E u |T .
Since g preserves a foliation without compact leaves, the integral matrix g∗ has an
eigenspace of irrational slope. This implies that either g∗ is hyperbolic or g∗ = Id. In
the second case, g has a lift bg : R2 → R2 such that b g = Id + α, where α is a periodic and,
in particular, a bounded function. Hence there exists a constant K > 0 such that, given any
X ⊂ R2 ,
diam(bg n (X )) ≤ diam(X ) + n K .
Let γ be an arc contained in a leaf of W u (x), x ∈ T . Then the length of γ grows
exponentially under ĝ n and its diameter grows at most linearly. This implies that, given a
small > 0, there exists an iterate of g that contains a curve of length arbitrarily large and
with endpoints at a distance less than . Using Poincaré–Bendixon we obtain a singularity
of the foliation W u . This is a contradiction and thus g∗ is hyperbolic.
Definition 6.5. Let f : M → M be a partially hyperbolic diffeomorphism. An embedded

torus tangential to E cs , E cu or E su will be called a cs, cu or a su-torus, respectively. In
these cases, we say that f admits the corresponding torus.
L EMMA 6.6. If f admits an su-torus, then M admits an Anosov torus.
Proof. Assume that f admits an su-torus and consider the lamination 3 of all su-tori of f .
This is a compact lamination [Hae62]. Therefore there is a recurrent leaf: that is, there is a
torus T and an iterate n such that d H ( f n (T ), T ) < ε for small ε. There exists a diffeotopy
i t on M taking f n (T ) into T . Then φ = f n ◦ i 1 fixes T and φ|T is isotopic to an Anosov
diffeomorphism.
L EMMA 6.7. If f admits an sc or cu torus, then it admits an f -periodic sc or a cu torus.

Therefore M admits an Anosov torus (by Proposition 6.4).
Proof. Let T be a cu-torus and consider the sequence f −n (T ). Since the family of all
compact subsets of M considered with the Hausdorff metric d H is compact, there is a
subsequence f −n k (T ) converging to a compact set K ⊂ M. Therefore, for each ε > 0,
there are arbitrarily large N >> L > 0 such that d H ( f −N (T ), f −L (T )) < ε.
Since T is transverse to the stable foliation, the union of all local stable leaves of T
forms a small tubular neighborhood of T , U (T ). Since stable leaves grow exponentially
under f −1 , if N >> L, as above, are large enough, then f −L (U (T )) ⊂ f −N (U (T )). This
implies that f N −L (U (T )) ⊂ U (T ).
Exercise 6.8. We finish the proof by showing that ∞ k(N −L) (U (T )) is a periodic
T
k=0 f
cu-torus.
C OROLLARY 6.9. Suppose that f : M → M is a partially hyperbolic diffeomorphism

admitting an sc, cu or su torus. If M is connected, then M is homeomorphic to:
(1) a 3-torus;
(2) the mapping torus of −Id : T2 → T2 ; or
(3) the mapping torus of a hyperbolic automorphism of T2 .
Observe that an sc or a cu torus cannot appear on the conservative setting. Indeed, by

Lemma 6.7 above, it would imply the existence of a periodic sc or cu torus. This 2-torus is,
respectively, repelling or attracting, a situation that cannot occur in a conservative setting.
In the following subsection, we sketch the proof of Theorem 6.2. We refer the reader
to [HHU11] for the complete proof.
6.1. Manifolds admitting Anosov tori. The first step in the proof of Theorem 6.2 is to
show that Anosov tori are incompressible.
Definition 6.10. An embedded orientable surface S ⊂ M is incompressible if the

homomorphism induced by the inclusion map i ∗ : π1 (S) → π1 (M) is injective.
Equivalently, S is incompressible if every embedded disc, D 2 ⊂ M, such that D 2 ∩ S =

∂ D 2 is contractible in S (see, for instance, [Hat07, p. 10]).
T HEOREM 6.11. (Hertz, Hertz and Ures [HHU08a]) Anosov tori are incompressible.
Now, let us assume, as in the hypotheses of Theorem 6.2, that the irreducible 3-manifold
M admits an Anosov torus T . Since T is incompressible, we can ‘cut’ M along T and
obtain a manifold N having incompressible 2-tori as boundary components. Theorem 6.2

then follows from the following theorem.
T HEOREM 6.12. (Hertz, Hertz and Ures [HHU11]) Let N be a compact orientable
irreducible 3-manifold with non-empty boundary such that all the boundary components
are incompressible 2-tori. If N admits an Anosov torus, then N ≈ T2 × [0, 1].
In order to prove this, we make use of the Jaco–Shalen–Johannson (JSJ) decomposition,
which states that any manifold satisfying the hypotheses of Theorem 6.12 can be cut
by a (unique) family of incompressible tori, so that the remaining pieces have certain
characteristics: they are either Seifert or else atoroidal and acylindrical. We provide these
definitions below (see also Theorem 6.15).
The proof of Theorem 6.12 consists of showing, on the one hand, that any Seifert
manifold having incompressible tori as boundary components is T2 × [0, 1] and, on the
other hand, that any manifold satisfying the hypotheses of Theorem 6.12 that is atoroidal
has an annulus which is properly embedded and is not isotopic to the boundary of the
manifold. This last statement contradicts the fact that the manifold is acylindrical, and
shows that every component in the JSJ-decomposition must be Seifert, which proves the
theorem.
Any compact 3-manifold, with or without boundary, supporting a foliation by circles is
a Seifert manifold (see [Eps81]). This was not the original definition; a more descriptive
one is the following.
Definition 6.13. A Seifert manifold is a 3-manifold that admits a decomposition into
disjoint circles, the fibers, such that each fiber has a neighborhood diffeomorphic,
preserving fibers, to either:
(1) a solid torus foliated by horizontal circles; or
(2) a solid torus foliated by the fibration obtained by the identification D 2 × [0, 1]/x ∼
R p/q (x), where R p/q : D 2 → D 2 denotes the rotation of angle p/q and p, q are
coprime.
If the manifold has boundary, its connected components are required to be tori, which are
also fibered by circles.
The circles of the first type are the generic fibers, while the ones of the second type are
the singular fibers. For an introduction to Seifert spaces see [Bri93].
Definition 6.14. Let N be a 3-manifold with boundary.
(1) N is atoroidal if every incompressible torus is ∂-parallel, that is, isotopic to a
subsurface of ∂ N .
(2) N is acylindrical if every incompressible annulus A that is properly embedded (that
is, ∂ A ⊂ ∂ N ) is ∂-parallel by an isotopy fixing ∂ A.
As we mentioned above, any irreducible orientable 3-manifold having incompressible
tori as boundary components admits a natural decomposition into Seifert pieces on one
side and atoroidal and acylindrical components on the other.
T HEOREM 6.15. (JSJ-decomposition [Hat07]) If N is an irreducible, orientable
3-manifold, having incompressible tori as boundary components, then there exists a finite
S
collection T of disjoint incompressible tori such that, for each component Ni of N \ T ,
either:
(1) Ni is a Seifert manifold; or
(2) Ni is both atoroidal and acylindrical.
Any minimal such collection is unique up to isotopy. This means that if T is a collection
as described above, it contains a minimal subcollection m(T ) satisfying the same claim.
All collections m(T ) are isotopic.
Any minimal family of incompressible tori, as described above, is called a JSJ-

decomposition of N . Note that if N is either atoroidal and acylindrical or Seifert, then
T = ∅.
Let us sketch how Theorem 6.12 is proved for the case of Seifert manifolds; the other
case is more delicate, and we refer the reader to [HHU11] for the complete proof.
Assume that N is a Siefert manifold, so it admits a foliation by circles, which is called
a Seifert fibration. We lose no generality in assuming that one of the incompressible tori
of the boundary of N is an Anosov torus T . By the definition of an Anosov torus, there
exists a diffeomorphism φ on N such that it is a linear hyperbolic automorphism of T . The
image of the Seifert fibration by φ is another Seifert fibration, which is non-isotopic to the
original one on T .
But orientable manifolds admitting two Seifert fibrations that are non-isotopic on its
boundary are completely classified.
L EMMA 6.16. [Hat07] If N admits two Seifert fibrations that are non-isotopic on ∂ N ,
then N is homeomorphic to:
(1) the solid torus;
(2) a twisted I bundle over the Klein bottle; or
(3) the torus cross the interval.
In the first two cases, ∂ N consists of a single torus, while, in last one, it consists of two
disjoint tori. To finish the proof, it suffices then to discard the first two cases.
L EMMA 6.17. If ∂ N contains an Anosov torus, then it contains more than one.
Proof. Assume that ∂ N is a torus T and consider the inclusion map i : H1 (∂ N ) → H1 (N ).

Let ker i be the kernel of the map. Then [Hat07, Lemma 3.5],
rank(ker i) = 12 rank(H1 (∂ N )),
where rank denotes the number of Z summands in a direct sum splitting into cyclic groups.
If ∂ M̃ = T , then (1/2)rank(H1 (T )) = 1, and hence K = ker i is a one-dimensional
subspace of H1 (T ). Then f ∗ (K ) = K , where f ∗ : H1 (T ) → H1 (T ) is the isomorphism
induced by any diffeomorphism f : N → N . This implies that f ∗ has 1 as an eigenvalue.
Hence f cannot be isotopic to a hyperbolic automorphism of T . This implies that T cannot
be an Anosov torus.
Acknowledgements. The authors want to thank Amie Wilkinson for her careful review
of this manuscript and for her many valuable suggestions that helped to improve its
readability. The first author is supported by FAPESP Project#2013/16226-8.
R EFERENCES
[ACW14] A. Avila, S. Crovisier and A. Wilkinson. Diffeomorphisms with positive metric entropy. Publ.
Math. Inst. Hautes Études Sci. 124(1) (2016), 319–347.
[Ano67] D. Anosov. Geodesic flows on Riemann manifolds with negative curvature. Proc. Steklov Inst. 90
(1967), 1–235.
[AS67] D. V. Anosov and Y. G. Sinai.. Certain smooth ergodic systems. Russian Math. Surveys 22 (1967),
103–167.
[Bar98] T. Barbot. Generalizations of the Bonatti–Langevin example of Anosov flow and their
classification up to topological equivalence. Comm. Anal. Geom. 6(4) (1998), 749–798.
[BBI04] M. Brin, D. Burago and S. Ivanov. On partially hyperbolic diffeomorphisms on 3-manifolds with
commutative fundamental group. Advances in Dynamical Systems. Cambridge University Press,
Cambridge, 2004, pp. 307–312.
[BBI09] M. Brin, D. Burago and S. Ivanov. Dynamical coherence of partially hyperbolic diffeomorphisms
of the 3-torus. J. Mod. Dyn. 3 (2009), 1–11.
[BDP02] K. Burns, D. Dolgopyat and Y. Pesin. Partial hyperbolicity, Lyapunov exponents and stable
ergodicity. J. Stat. Phys. 108 (2002), 927–942. Dedicated to David Ruelle and Yasha Sinai on
the occasion of their 65th birthdays.
[BGP15] C. Bonatti, A. Gogolev and R. Potrie. Anomalous partially hyperbolic diffeomorphisms II: stably
ergodic examples. Invent. Math. 206(3) (2016), 801–836.
[BI08] D. Burago and S. Ivanov. Partially hyperbolic diffeomorphisms of 3-manifolds with abelian
fundamental groups. J. Mod. Dyn. 2 (2008), 541–580.
[BL94] C. Bonatti and R. Langevin. Un exemple de flot d’Anosov transitif transverse à un tore et non
conjugué à une suspension. Ergod. Th. & Dynam. Sys. 14(4) (1994), 633–643.
[BP74] M. Brin and Y. Pesin. Partially hyperbolic dynamical systems. Izv. Akad. Nauk SSSR Ser. Mat. 38
(1974), 170–212.
[BPP14] C. Bonatti, K. Parwani and R. Potrie. Anomalous partially hyperbolic diffeomorphisms I:
dynamically coherent examples. Ann. Sci. Éc. Norm. Supér., to appear, Preprint, 2014,
arXiv:1411.1221v1.
[Bri93] M. Brin. Seifert fibered spaces. Preprint, 1993, arXiv:0711.1346.
[Bru93] M. Brunella. Expansive flows on seifert manifolds and on torus bundles. Bull. Braz. Math. Soc.
(N.S.) 241 (1993), 89–104.
[BW05a] C. Bonatti and A. Wilkinson. Transitive partially hyperbolic diffeomorphisms on 3-manifolds.
Topology 44(3) (2005), 475–508.
[BW05b] K. Burns and A. Wilkinson. Better center bunching. Preprint, 2005.
[BW08] K. Burns and A. Wilkinson. Dynamical coherence and center bunching. Discrete Contin. Dyn.
Syst. 22(1–2) (2008), 89–100.
[BW10] K. Burns and A. Wilkinson. On the ergodicity of partially hyperbolic systems. Ann. of Math. (2)
171 (2010), 451–489.
[Did03] P. Didier. Stability of accessibility. Ergod. Th. & Dynam. Sys. 23(6) (2003), 1717–1731.
[DW03] D. Dolgopyat and A. Wilkinson. Stable accessibility is C 1 dense. Astérisque 287 (2003.), 33–60.
[Eps81] D. B. A. Epstein. Pointwise periodic homeomorphisms. Proc. Lond. Math. Soc. (3) s3-42(3)
(1981), 415–460.
[Fen94] S. Fenley. Anosov flows in 3-manifolds. Ann. of Math. (2) 139(1) (1994), 79–115.
[Fri83] D. Fried. Transitive anosov flows and pseudo-anosov maps. Topology 22(3) (1983), 299–303.
[FW80a] J. Franks and R. Williams. Anomalous Anosov flows. Global Theory of Dynamical Systems (Proc.
Int. Conf., Northwestern University, Evanston, IL, 1979 (Lecture Notes in Mathematics, 819).
Springer, Berlin, 1980, pp. 158–174.
[FW80b] J. Franks and B. Williams. Anomalous Anosov flows. Global Theory of Dynamical Systems (Proc.
Int. Conf., Northwestern University, Evanston, IL, 1979 (Lecture Notes in Mathematics, 819).
Springer, Berlin, 1980, pp. 158–174.
[GPS94] M. Grayson, C. Pugh and M. Shub. Stably ergodic diffeomorphisms. Ann. of Math. (2) 140 (1994),
295–329.
[Hae62] A. Haefliger. Variétés feuilletées. Topologia Differenziale (Centro Internaz. Mat. Estivo, 1 deg
Ciclo, Urbino, 1962), Lezione 2. Edizioni Cremonese, Rome, 1962, p. 46.
[Ham13] A. Hammerlindl. Leaf conjugacies on the torus. Ergod. Th. & Dynam. Sys. 3(33) (2013), 896–933.
[Hat07] A. Hatcher. Notes on basic 3-manifold topology. Available at: www.math.cornell.edu/ hatcher/,
2007.
[HHTU11] F. Rodriguez Hertz, M. A. Rodriguez Hertz, A. Tahzibi and R. Ures. New criteria for ergodicity
and nonuniform hyperbolicity. Duke Math. J. 160(3) (2011), 599–629.
[HHU08a] F. Rodriguez Hertz, M. A. Rodriguez Hertz and R. Ures. Partial hyperbolicity and ergodicity in
dimension three. J. Mod. Dyn. 2 (2008), 187–208.
[HHU08b] F. Rodriguez Hertz, M. Rodriguez Hertz and R. Ures. Accessibility and stable ergodicity
for partially hyperbolic diffeomorphisms with 1d-center bundle. Invent. Math. 2(172) (2008),
353–381.
[HHU08c] F. Rodriguez Hertz, M. Rodriguez Hertz and R. Ures. On the existence and uniqueness of weak
foliations in dimension 3. Geom. Probab. Struct. Dyn. 469 (2008), 303–316.
[HHU11] F. Rodriguez Hertz, M. Rodriguez Hertz and R. Ures. Tori with hyperbolic dynamics in
3-manifolds. J. Mod. Dyn. 5(1) (2011), 185–202.
[HHU15a] F. Rodriguez Hertz, M. Rodriguez Hertz and R. Ures. Center-unstable foliations do not have
compact leaves. Math. Res. Lett. (2015), to appear.
[HHU15b] F. Rodriguez Hertz, M. A. Rodriguez Hertz and R. Ures. A non-dynamically coherent example on
T3 . Ann. Inst. Henri-Poincaré 33(4) (2016), 1023–1032.
[HP13] A. Hammerlindl and R. Potrie. Classification of partially hyperbolic diffeomorphisms in
3-manifolds with solvable fundamental group. J. Topol. 8(3) (2015), 842–870.
[HP14] A. Hammerlindl and R. Potrie. Pointwise partial hyperbolicity in 3-dimensional nilmanifolds.
J. Lond. Math. Soc. (2) 3(89) (2014), 853–875.
[HP15] A. Hammerlindl and R. Potrie. Partial hyperbolicity and classification: a survey. Ergod. Th. &
Dynam. Sys., to appear, Preprint, 2015, arXiv:1511.04471.
[HPS77] M. Hirsch, C. Pugh and M. Shub. Invariant Manifolds. Springer, Berlin, 1977.
[HT80] M. Handel and W. P. Thurston. Anosov flows on new three manifolds. Invent. Math. 59(2) (1980),
95–103.
[Jou88] J.-L. Journé. A regularity lemma for functions of several variables. Rev. Mat. Iberoam. 4 (1988),
187–193.
[KH95] A. Katok and B. Hasselblatt. Introduction to the Modern Theory of Dynamical Systems
(Encyclopedia of Mathematics and its Applications, 54). Cambridge University Press, Cambridge,
1995.
[Mañ78] R. Mañé. Contributions to the stability conjecture. Topology 17 (1978), 383–396.
[MS12] C. G. Moreira and W. Silva. On the geometry of horseshoes in higher dimensions. Preprint, 2012,
arXiv:1210.2623.
[Pot12] R. Potrie. Partial hyperbolicity and foliations in ≈3 . J. Mod. Dyn. 9(1) (2015), 81–121.
[PS72] C. Pugh and M. Shub. Ergodicity of anosov actions. Invent. Math. 15 (1972), 1–23.
[PS96] C. Pugh and M. Shub. Stable ergodicity and partial hyperbolicity. Proc. 1st Int. Conf. on
Dynamical Systems (Montevideo, Uruguay, 1995) (Research Notes Mathematics Series, 362). Eds.
S. Newhouse, F. Ledrappier and J. Lewowicz. Longman, Pitman, Harlow, 1996, pp. 182–187.
[PS97] C. Pugh and M. Shub.. Stably ergodic dynamical systems and partial hyperbolicity. J. Complexity
13 (1997), 125–179.
[PS00] C. Pugh and M. Shub. Stable ergodicity and julienne quasiconformality. J. Eur. Math. Soc. (JEMS)
2(1) (2000), 1–52.
[PSW97] C. Pugh, M. Shub and A. Wilkinson. Holder foliations. Duke Math. J. 86(3) (1997), 517–546.
[Ros68] H. Rosenberg. Foliations by planes. Topology 7(2) (1968), 131–138.
[Rou71] R. Roussarie. Sur les feuilletages des variétés de dimension trois. Ann. Inst. Fourier (Grenoble) 21
(1971), 13–82.
[Sac70] R. Sacksteder. Strongly mixing transformations. Global Analysis (Proc. Sympos. Pure Math., Vol.
XIV, Berkeley, CA, 1968). American. Mathematical Society, Providence, RI, 1970, pp. 245–252.
[Sma67] S. Smale. Differentiable dynamical systems. Bull. Amer. Math. Soc. (N.S.) 73 (1967), 747–817.
[Wil98] A. Wilkinson. Stable ergodicity of the time-one map of a geodesic flow. Ergod. Th. & Dynam. Sys.
18(6) (1998), 1545–1587.

Partially Hyperbolic Dynamics in Dimension Three PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Partially Hyperbolic Dynamics in Dimension Three PDF

Hochgeladen von

Copyright:

Verfügbare Formate

c Cambridge University Press, 2017

Ergod. Th. & Dynam. Sys. (2018), 38, 2801–2837

(Received 5 December 2016 and accepted in revised form 5 December 2016)

Abstract. Partial hyperbolicity appeared in the 1960s as a natural generalization of

simplest examples are certain automorphisms of the torus, for instance

ψ, respectively, contracts, expands and is an isometry on the leaves of these foliations.

s ), ψ(g)) = d(gd1 u e±1 s , gd1 ) = d(u e±1 s , e) = d(gu e±1 s , g)

shows that ψ acts isometrically on the leaves of F c . Define E gσ = Tg F σ , σ = s, u, c and

[D, U + ] = U + , [D, U − ] = −U − , [U + , U − ] = 2D. (1.4)

Let E eu = span{U + }, E es = span{U − } and E ec = span{D}. Clearly, U + , U − , D are

Given g ∈ G, let us define Rg : G → G by Rg (h) = hg. Let De Rg : Te G → Tg G be the

Precisely, we say that a diffeomorphism f of a closed manifold M is partially

kD f (x)v s k < 1 < kD f (x)v u k and (1.5)

1.1.1. Time-one maps of Anosov flows. Consider an Anosov flow in a 3-manifold φt :

1.1.2. Skew products. Another example of a partially hyperbolic diffeomorphism is

1.1.3. DA-diffeomorphisms. A DA (‘Derived from Anosov’) partially hyperbolic

2. Some open problems

F IGURE 1. Mañé’s DA partially hyperbolic diffeomorphism.

2.1. Ergodicity. As stated above, one problem in partially hyperbolic dynamics

Exercise 2.5. Show that the automorphism on T3 defined by

T HEOREM 2.12. [HHU11] If a partially hyperbolic diffeomorphism f : M 3 → M 3 has

The following problem is also open.

Problem 2.14. Let f be a conservative non-ergodic partially hyperbolic diffeomorphism

2.2. Dynamical coherence. In partially hyperbolic dynamics, the strong bundles E s

2.2.1. A non-dynamically coherent example. The aforementioned example is, in fact,

Definition 2.15. A partially hyperbolic diffeomorphism is dynamically coherent if the

kD f (x)v s k ≤ λ ≤ kD f (x)v c k ≤ µ ≤ kD f (x)v u k. (2.8)

In [HHU15a], it was shown that if there is an invariant foliation tangential to the

(1) the 3-torus T3 ;

Question 2.24. Is it possible to classify partially hyperbolic dynamics in 3-manifolds,

is almost everywhere constant. But ϕ̃(x) coincides almost everywhere with

path consisting of a concatenation of stable and unstable leaves would be contained in

This implies that

T HEOREM 3.3. [BW10, HHU08b] If f ∈ Diff2m (M 3 ) is partially hyperbolic and satisfies

In fact, Burns and Wilkinson proved a much more general result.

T HEOREM 3.4. [BW10] If f is a partially hyperbolic diffeomorphism (with any center

A diffeomorphism is said to satisfy the center bunching property if

where m(T ) = kT −1 k−1 .

F IGURE 3. An su-path from y to z

3.2. Properties of accessibility classes. We want to describe precisely non-ergodic

Let us begin with a local description of open accessibility classes.

F IGURE 4. An open accessibility class.

F IGURE 5. An accessibility class in 0( f ).

to E u . This implies that W u (D sc ) is contained in AC(x) and contains an open set.

accessibility classes of points in 0( f ).

Hence, due to Lemma 3.7 above, for each x ∈ 0( f ),

See Figure 5 for an illustration of a point of joint integrability of W s and W u .

P ROPOSITION 3.9. A point x belongs to 0( f ) if and only if W s and W u are jointly

where δ > 0 is as in the definition of joint integrability (Definition 3.8), is a two-

be seen as the graph of a continuous function η : D → N .

Since π is smooth, we also obtain that F h = {π(Wloc s (η(v)))}

continuous in the C 1 -topology.

3.3. Properties of the lamination 0( f ) of non-open accessibility classes. In order to

T HEOREM 3.12. [HHU08a, Theorem 1.6] Let f : M 3 → M 3 be a conservative partially

With respect to item (2), a leaf L of a lamination 3 is a boundary leaf if there is

F IGURE 6. Reeb component.

T HEOREM 3.18. (Hertz, Hertz and Ures) If 3 ⊂ 0( f ) is an orientable and transversely

P ROPOSITION 4.1. If f : M 3 → M 3 is a partially hyperbolic diffeomorphism whose

Proof. First, observe that W c is f -invariant: if c : [0, 1] → W c (x) is a differentiable

saturated by the corresponding strong foliations. Again using uniqueness of solutions

Problem 4.2. Find an example of a dynamically coherent partially hyperbolic

Another condition that guarantees dynamical coherence in T3 is absolute partial

T HEOREM 4.3. (Brin–Burago–Ivanov [BBI09]) If f : T3 → T3 is absolutely partially